feat: add realistic portable localization core

packages/acoustics-core/README.md

@@ -3,11 +3,56 @@
 두 앱이 공유하는 무의존성 수치 기준 구현입니다.
 
 - 결정적 tone/white/pink noise 및 오디오 asset resampling
-- 자유음장 점 음원의 `delay = r / c`, `amplitude = 1 / r` 구면 전파
+- 자유음장 직접음과 선택적 저비용 shoebox multipath
-- zero-padded FFT phase shift 기반 fractional sample delay
+- zero-padded FFT 전달함수 기반 fractional sample delay와 마이크 주파수 응답
 - radix-2 FFT, Hann STFT와 선택 bin 추출
 - scan grid 거리 cache 및 conventional frequency-domain DAS
 
+## 실측 다채널 스트림 위치 추정
+
+`realtime-localizer.js`는 Scene·음원 합성 없이 이미 sampling된 동기식 마이크 채널을 직접
+받습니다. 외부 package 의존성이 없는 ES module이며, 다른 프로젝트에는 `acoustics-core`
+디렉터리를 그대로 복사해 사용할 수 있습니다. 채널 순서는 `array.microphones` 순서와 같아야 하고
+각 chunk는 동일 길이의 `Float32Array` 목록이어야 합니다.
+
+```js
+import { createStreamingLocalizer } from "./packages/acoustics-core/realtime-localizer.js";
+
+const localizer = createStreamingLocalizer({
+  sampleRate: 48_000,
+  array,
+  scan: { xMin: -1, xMax: 1, yMin: -1, yMax: 1, z: 2, columns: 50, rows: 50 },
+  bands: [{ id: "machine", minHz: 900, maxHz: 1100, color: "#65f2bf" }],
+  stft: { fftSize: 512, hopSize: 256 },
+  analysisWindowSamples: 4_800,
+  analysisHopSamples: 4_800,
+  maximumBufferedSamples: 19_200,
+  maximumResultsPerCall: 1,
+});
+
+// Capture callback: 복사 후 즉시 반환하며, 오래 밀린 sample은 oldest-first로 버립니다.
+localizer.append(channels);
+
+// Analysis loop: 호출마다 설정된 상한 이하의 map만 반환합니다.
+const results = await localizer.analyzeAvailable({ signal: abortController.signal });
+for (const result of results) {
+  const { position, rawPower } = result.locations[0];
+  drawMarker(position, rawPower, result.centerTime);
+}
+```
+
+한 블록만 처리할 때는 `createLocalizationPlan()`과 `localizeMultichannel()`을 사용합니다.
+`getState()`은 absolute sample origin, buffered/dropped sample, 대기 window 수와 고정 input
+buffer byte 수를 돌려주므로 capture adapter가 지연을 감시할 수 있습니다. 실제 장치의 clock,
+channel framing과 transport는 하드웨어별 adapter 책임이며 이 모듈은 그 다음 단계부터 시작합니다.
+
+반복 CPU 분석 계획은 거리 행렬, bin calibration, 제한 내의 Float64 steering matrix와 임시 bin map을
+재사용합니다. 예를 들어 192채널·50×50·2 bins를 10회 분석하면 거리 계산은 4,800,000회에서
+480,000회로, steering point/microphone pair는 9,600,000회에서 960,000회로 줄어듭니다
+(sin/cos 호출 17,280,000회 회피). 이 경우 약 16.5 MiB를 유지하며, 기본 64 MiB steering 한도를
+넘는 설정은 steering cache만 끕니다. 자동 테스트는 cache 전후 raw/display Float32 map의 전 요소
+bit-exact 일치를 고정합니다.
+
 ## 이동 음원과 분석 시간창
 
 `sourcePositionAtTime`은 scene 시작 기준의 절대 시각에서 trajectory keyframe 사이를 선형 보간하며,
@@ -33,8 +78,40 @@ calibration 값의 역 gain/phase로 보정합니다. raw power map과 표시용
 buffer로 보존합니다. bin spectrum은 제한된 batch로 처리하며 point batch 사이에서 event loop를
 양보해 진행 callback과 `AbortSignal` 취소가 동작합니다.
 
-현재 전파 모델은 자유음장·무지향성 점 음원이며 잔향은 포함하지 않습니다. 전체 signal을
+마이크별 `frequencyResponse`는 2~64개의 `{ frequencyHz, gainDb, phaseDeg? }` 점으로 정의합니다.
-메모리에 보관하는 CPU 기준 구현이므로 source/channel과 delay FFT working set을 512 MiB로 제한합니다.
+`interpolateFrequencyResponse`는 dB와 wrapped phase를 구간 선형 보간하고 측정 범위 밖에서는 가장
-장시간 입력은 이후 chunk/STFT streaming backend에서 지원합니다.
+가까운 끝점을 사용합니다. 합성 시 기존 sensitivity/phase 편차에 이 응답을 추가하며, CPU DAS는
+선택 FFT bin의 역응답을 calibration/steering cache에 포함합니다. WebGPU 경로는 GPU 전송 직전의
+packed STFT에 같은 역응답을 적용합니다. `beamforming.compensateFrequencyResponse: false`는
+sensitivity/phase 보정은 유지한 채 곡선 보정만 꺼 robustness 비교를 재현합니다.
+
+응답 곡선이 없는 배열은 기존 계산 경로와 결과를 유지합니다. 곡선이 있으면 합성 FFT마다 bin당
+보간·복소 곱이 추가되고, WebGPU 준비 단계에는 `선택 bin × 마이크` calibration과 STFT 복소 곱이
+추가되지만 GPU buffer layout이나 dispatch 크기는 늘어나지 않습니다.
+
+## 저비용 환경 근사
+
+Scene의 `acoustics`를 생략하거나 `{ "model": "free-field" }`로 두면 기존 직접음
+`delay = r/c`, `gain = 1/r`만 사용합니다. `model: "shoebox"`는 다음 경로를 선택적으로
+추가합니다.
+
+- 직육면체 `room`의 여섯 면에 대한 1차 image-source 반사
+- 반사 에너지 일부를 deterministic 지연 경로로 나누는 scattering 근사
+- 축 정렬 box `occluders`를 통과하는 직접음 감쇠와 가장 짧은 corner detour 회절 경로
+- RT60·gain·tap 수로 제한한 deterministic diffuse-reverb tail
+
+모든 경로의 위상과 gain은 마이크마다 하나의 frequency-domain 전달함수로 합친 뒤 IFFT를 한 번만
+수행합니다. 경로 위상은 recurrence로 갱신하고 1,024 bin마다 재고정하므로 각 bin·path마다
+삼각함수를 다시 호출하지 않습니다. UI preset도 수신점당 sparse path 상한을 표시합니다.
+
+이 모델은 배열 배치와 위치추정 robustness를 빠르게 비교하기 위한 근사입니다. 파동 방정식,
+고차 반사, 재질별 주파수 의존 흡음, 복잡한 mesh 회절, Doppler, 음원·마이크 지향성, 압력/SPL,
+센서 self-noise, ADC quantization·clock jitter는 계산하지 않습니다. 그런 효과가 중요한
+검증에서는 실측 다채널 데이터를 아래 streaming API에 넣어 결과를 비교해야 합니다.
+
+snapshot 합성은 전체 signal과 가장 늦은 sparse path를 포함하는 delay FFT working set을
+512 MiB로 제한합니다. 실측·장시간 입력은 위 streaming localizer의 bounded ring buffer로
+전체 녹음을 메모리에 보관하지 않고 처리할 수 있습니다. CPU analysis plan도 기본 512 MiB
+hard cap을 할당 전에 검사하며 steering cache는 별도의 64 MiB 한도를 넘으면 자동으로 끕니다.
 
 192채널·50×50 fixture는 브라우저 통합 테스트에서 실제 합성부터 DAS peak 검출까지 실행합니다.

packages/acoustics-core/das.js

@@ -1,4 +1,5 @@
 import { dbToLinear, createScanPoints, distance3, findPeak, scanIndexToPosition, TAU } from "./math.js";
+import { interpolateFrequencyResponse } from "./frequency-response.js";
 import { binsForBand, extractStftBins, getFrameCount } from "./stft.js";
 
 function throwIfAborted(signal) {
@@ -13,6 +14,10 @@ function yieldToEventLoop() {
   return new Promise((resolve) => setTimeout(resolve, 0));
 }
 
+export const DEFAULT_MAXIMUM_CPU_STEERING_CACHE_BYTES = 64 * 1024 * 1024;
+export const DEFAULT_MAXIMUM_CPU_ANALYSIS_PLAN_BYTES = 512 * 1024 * 1024;
+const DAS_ANALYSIS_PLAN_KIND = "micarrsim.das-analysis-plan/1";
+
 export function createDistanceMatrix(arrayDefinition, scan) {
   const points = createScanPoints(scan);
   const microphoneCount = arrayDefinition.microphones.length;
@@ -31,6 +36,94 @@ export function createDistanceMatrix(arrayDefinition, scan) {
   return distances;
 }
 
+function createBandIndexLookup(bandBins) {
+  const lookup = new Map();
+  bandBins.forEach((bins, bandIndex) => {
+    bins.forEach((bin) => {
+      if (!lookup.has(bin)) lookup.set(bin, []);
+      lookup.get(bin).push(bandIndex);
+    });
+  });
+  return lookup;
+}
+
+function createBinCalibration(arrayDefinition, bin, sampleRate, fftSize, compensateFrequencyResponse) {
+  const microphoneCount = arrayDefinition.microphones.length;
+  const gains = new Float64Array(microphoneCount);
+  const phases = new Float64Array(microphoneCount);
+  const frequencyHz = (bin * sampleRate) / fftSize;
+  const response = {};
+  for (let microphoneIndex = 0; microphoneIndex < microphoneCount; microphoneIndex += 1) {
+    const microphone = arrayDefinition.microphones[microphoneIndex];
+    let responseGainDb = 0;
+    let responsePhaseDeg = 0;
+    if (compensateFrequencyResponse && microphone.frequencyResponse) {
+      interpolateFrequencyResponse(microphone.frequencyResponse, frequencyHz, response);
+      responseGainDb = response.gainDb;
+      responsePhaseDeg = response.phaseDeg;
+    }
+    gains[microphoneIndex] = dbToLinear(
+      -(microphone.sensitivityDb ?? 0) - responseGainDb,
+    );
+    phases[microphoneIndex] =
+      ((microphone.phaseOffsetDeg ?? 0) + responsePhaseDeg) * Math.PI / 180;
+  }
+  return { gains, phases };
+}
+
+function fillSteeringVector({
+  angularWaveNumber,
+  distances,
+  distanceOffset,
+  calibration,
+  outputReal,
+  outputImaginary,
+  outputOffset,
+}) {
+  for (let microphoneIndex = 0; microphoneIndex < calibration.gains.length; microphoneIndex += 1) {
+    const phase =
+      angularWaveNumber * distances[distanceOffset + microphoneIndex] -
+      calibration.phases[microphoneIndex];
+    outputReal[outputOffset + microphoneIndex] =
+      Math.cos(phase) * calibration.gains[microphoneIndex];
+    outputImaginary[outputOffset + microphoneIndex] =
+      Math.sin(phase) * calibration.gains[microphoneIndex];
+  }
+}
+
+async function createSteeringMatrix(options) {
+  const {
+    bin,
+    pointCount,
+    microphoneCount,
+    distances,
+    sampleRate,
+    fftSize,
+    speedOfSound,
+    calibration,
+    signal,
+    pointBatchSize,
+  } = options;
+  const real = new Float64Array(pointCount * microphoneCount);
+  const imaginary = new Float64Array(pointCount * microphoneCount);
+  const angularWaveNumber = (TAU * bin * sampleRate) / (fftSize * speedOfSound);
+  for (let pointIndex = 0; pointIndex < pointCount; pointIndex += 1) {
+    throwIfAborted(signal);
+    const offset = pointIndex * microphoneCount;
+    fillSteeringVector({
+      angularWaveNumber,
+      distances,
+      distanceOffset: offset,
+      calibration,
+      outputReal: real,
+      outputImaginary: imaginary,
+      outputOffset: offset,
+    });
+    if ((pointIndex + 1) % pointBatchSize === 0) await yieldToEventLoop();
+  }
+  return { real, imaginary };
+}
+
 export async function computeBinMap(options) {
   const {
     bin,
@@ -42,26 +135,64 @@ export async function computeBinMap(options) {
     sampleRate,
     fftSize,
     speedOfSound,
+    compensateFrequencyResponse = true,
+    calibration: providedCalibration,
+    steeringMatrix,
+    output: providedOutput,
+    steeringReal: providedSteeringReal,
+    steeringImaginary: providedSteeringImaginary,
     signal,
-    pointBatchSize,
+    pointBatchSize = 64,
   } = options;
   const microphoneCount = arrayDefinition.microphones.length;
-  const output = new Float32Array(pointCount);
+  const output = providedOutput ?? new Float32Array(pointCount);
-  const steeringReal = new Float64Array(microphoneCount);
+  const steeringReal = providedSteeringReal ?? new Float64Array(microphoneCount);
-  const steeringImaginary = new Float64Array(microphoneCount);
+  const steeringImaginary = providedSteeringImaginary ?? new Float64Array(microphoneCount);
   const angularWaveNumber = (TAU * bin * sampleRate) / (fftSize * speedOfSound);
   const microphoneScale = 1 / microphoneCount;
+  const calibration = providedCalibration ?? createBinCalibration(
+    arrayDefinition,
+    bin,
+    sampleRate,
+    fftSize,
+    compensateFrequencyResponse,
+  );
+  const phaseRealValues = steeringMatrix?.real ?? steeringReal;
+  const phaseImaginaryValues = steeringMatrix?.imaginary ?? steeringImaginary;
+  if (output.length !== pointCount) throw new RangeError("output length must match pointCount.");
+  if (steeringReal.length !== microphoneCount || steeringImaginary.length !== microphoneCount) {
+    throw new RangeError("steering workspace length must match microphone count.");
+  }
+  if (
+    calibration.gains.length !== microphoneCount ||
+    calibration.phases.length !== microphoneCount
+  ) {
+    throw new RangeError("calibration length must match microphone count.");
+  }
+  if (
+    steeringMatrix &&
+    (
+      steeringMatrix.real.length !== pointCount * microphoneCount ||
+      steeringMatrix.imaginary.length !== pointCount * microphoneCount
+    )
+  ) {
+    throw new RangeError("steering matrix length must match point and microphone counts.");
+  }
 
   for (let pointIndex = 0; pointIndex < pointCount; pointIndex += 1) {
     throwIfAborted(signal);
     const distanceOffset = pointIndex * microphoneCount;
-    for (let microphoneIndex = 0; microphoneIndex < microphoneCount; microphoneIndex += 1) {
+    const steeringOffset = steeringMatrix ? distanceOffset : 0;
-      const microphone = arrayDefinition.microphones[microphoneIndex];
+    if (!steeringMatrix) {
-      const calibrationPhase = ((microphone.phaseOffsetDeg ?? 0) * Math.PI) / 180;
+      fillSteeringVector({
-      const phase = angularWaveNumber * distances[distanceOffset + microphoneIndex] - calibrationPhase;
+        angularWaveNumber,
-      const calibrationGain = dbToLinear(-(microphone.sensitivityDb ?? 0));
+        distances,
-      steeringReal[microphoneIndex] = Math.cos(phase) * calibrationGain;
+        distanceOffset,
-      steeringImaginary[microphoneIndex] = Math.sin(phase) * calibrationGain;
+        calibration,
+        outputReal: steeringReal,
+        outputImaginary: steeringImaginary,
+        outputOffset: 0,
+      });
     }
 
     let accumulatedPower = 0;
@@ -73,8 +204,8 @@ export async function computeBinMap(options) {
         const spectrumIndex = spectrumOffset + microphoneIndex;
         const inputReal = spectrum.real[spectrumIndex];
         const inputImaginary = spectrum.imaginary[spectrumIndex];
-        const phaseReal = steeringReal[microphoneIndex];
+        const phaseReal = phaseRealValues[steeringOffset + microphoneIndex];
-        const phaseImaginary = steeringImaginary[microphoneIndex];
+        const phaseImaginary = phaseImaginaryValues[steeringOffset + microphoneIndex];
         sumReal += inputReal * phaseReal - inputImaginary * phaseImaginary;
         sumImaginary += inputReal * phaseImaginary + inputImaginary * phaseReal;
       }
@@ -86,11 +217,200 @@ export async function computeBinMap(options) {
   return output;
 }
 
-export function estimateDasWorkload(options) {
+/**
-  const { channels, sampleRate, arrayDefinition, experiment } = options;
+ * Cache geometry and optional steering matrices for repeated CPU DAS analyses.
-  const { fftSize, hopSize } = experiment.stft;
+ *
+ * Plans are intentionally stateful and must not be used concurrently. They are
+ * local runtime objects (Maps and workspaces), not serializable data contracts.
+ */
+export function createDasAnalysisPlan(options) {
+  const {
+    sampleRate,
+    arrayDefinition,
+    scene,
+    experiment,
+    maximumSteeringCacheBytes = DEFAULT_MAXIMUM_CPU_STEERING_CACHE_BYTES,
+    maximumPlanBytes = DEFAULT_MAXIMUM_CPU_ANALYSIS_PLAN_BYTES,
+  } = options;
+  if (!Number.isFinite(sampleRate) || sampleRate <= 0) {
+    throw new RangeError("sampleRate must be a positive finite number.");
+  }
+  if (!Array.isArray(arrayDefinition?.microphones) || arrayDefinition.microphones.length === 0) {
+    throw new TypeError("arrayDefinition.microphones must be a non-empty array.");
+  }
+  if (!Number.isFinite(scene?.speedOfSound) || scene.speedOfSound <= 0) {
+    throw new RangeError("scene.speedOfSound must be a positive finite number.");
+  }
+  if (!Number.isFinite(maximumSteeringCacheBytes) || maximumSteeringCacheBytes < 0) {
+    throw new RangeError("maximumSteeringCacheBytes must be a non-negative finite number.");
+  }
+  if (!Number.isSafeInteger(maximumPlanBytes) || maximumPlanBytes <= 0) {
+    throw new RangeError("maximumPlanBytes must be a positive safe integer.");
+  }
+
+  const { fftSize } = experiment.stft;
   const bandBins = experiment.frequencyBands.map((band) => binsForBand(band, sampleRate, fftSize));
   const selectedBins = [...new Set(bandBins.flat())].sort((a, b) => a - b);
+  const microphoneCount = arrayDefinition.microphones.length;
+  const pointCount = experiment.scan.columns * experiment.scan.rows;
+  const pointMicrophonePairs = pointCount * microphoneCount;
+  const distanceMatrixBytes = pointMicrophonePairs * Float32Array.BYTES_PER_ELEMENT;
+  const retainedBaseBytes =
+    distanceMatrixBytes +
+    pointCount * Float32Array.BYTES_PER_ELEMENT +
+    microphoneCount * Float64Array.BYTES_PER_ELEMENT * 2;
+  if (
+    !Number.isSafeInteger(pointMicrophonePairs) ||
+    !Number.isSafeInteger(retainedBaseBytes) ||
+    retainedBaseBytes > maximumPlanBytes
+  ) {
+    throw new RangeError(
+      "CPU analysis plan requires " + retainedBaseBytes.toLocaleString() +
+      " base bytes, above maximumPlanBytes " + maximumPlanBytes.toLocaleString() + ".",
+    );
+  }
+  const distances = createDistanceMatrix(arrayDefinition, experiment.scan);
+  const binMap = new Float32Array(pointCount);
+  const steeringReal = new Float64Array(microphoneCount);
+  const steeringImaginary = new Float64Array(microphoneCount);
+  const steeringBytesPerBin =
+    pointCount * microphoneCount * Float64Array.BYTES_PER_ELEMENT * 2;
+  const requestedSteeringCacheBytes = steeringBytesPerBin * selectedBins.length;
+  const steeringCacheEnabled =
+    Number.isSafeInteger(requestedSteeringCacheBytes) &&
+    requestedSteeringCacheBytes > 0 &&
+    requestedSteeringCacheBytes <= maximumSteeringCacheBytes &&
+    retainedBaseBytes + requestedSteeringCacheBytes <= maximumPlanBytes;
+
+  return {
+    kind: DAS_ANALYSIS_PLAN_KIND,
+    sampleRate,
+    arrayDefinition,
+    scene,
+    experiment,
+    microphoneCount,
+    pointCount,
+    bandBins,
+    selectedBins,
+    distances,
+    bandIndexesByBin: createBandIndexLookup(bandBins),
+    calibrations: new Map(),
+    steeringMatrices: new Map(),
+    steeringCacheEnabled,
+    steeringBytesPerBin,
+    requestedSteeringCacheBytes,
+    maximumSteeringCacheBytes,
+    maximumPlanBytes,
+    binMap,
+    steeringReal,
+    steeringImaginary,
+    busy: false,
+    statistics: {
+      analysisAttempts: 0,
+      completedAnalyses: 0,
+      distanceMatrixBuilds: 1,
+      distanceEvaluations: pointCount * microphoneCount,
+      avoidedDistanceEvaluations: 0,
+      calibrationBuilds: 0,
+      calibrationCacheHits: 0,
+      steeringMatrixBuilds: 0,
+      steeringCacheHits: 0,
+      steeringPointMicrophonePairs: 0,
+      avoidedSteeringPointMicrophonePairs: 0,
+      binMapUses: 0,
+      avoidedBinMapAllocations: 0,
+      retainedBytes: retainedBaseBytes,
+    },
+  };
+}
+
+function assertCompatibleAnalysisPlan(plan, options) {
+  if (plan?.kind !== DAS_ANALYSIS_PLAN_KIND) {
+    throw new TypeError("analysisPlan was not created by createDasAnalysisPlan().");
+  }
+  if (
+    plan.sampleRate !== options.sampleRate ||
+    plan.arrayDefinition !== options.arrayDefinition ||
+    plan.scene !== options.scene ||
+    plan.experiment !== options.experiment
+  ) {
+    throw new Error("analysisPlan must be used with the exact configuration objects that created it.");
+  }
+}
+
+function planCalibration(plan, bin) {
+  let calibration = plan.calibrations.get(bin);
+  if (calibration) {
+    plan.statistics.calibrationCacheHits += 1;
+    return calibration;
+  }
+  calibration = createBinCalibration(
+    plan.arrayDefinition,
+    bin,
+    plan.sampleRate,
+    plan.experiment.stft.fftSize,
+    plan.experiment.beamforming.compensateFrequencyResponse !== false,
+  );
+  plan.calibrations.set(bin, calibration);
+  plan.statistics.calibrationBuilds += 1;
+  plan.statistics.retainedBytes += calibration.gains.byteLength + calibration.phases.byteLength;
+  return calibration;
+}
+
+async function planSteeringMatrix(plan, bin, calibration, signal, pointBatchSize) {
+  if (!plan.steeringCacheEnabled) return undefined;
+  let matrix = plan.steeringMatrices.get(bin);
+  if (matrix) {
+    plan.statistics.steeringCacheHits += 1;
+    plan.statistics.avoidedSteeringPointMicrophonePairs +=
+      plan.pointCount * plan.microphoneCount;
+    return matrix;
+  }
+  matrix = await createSteeringMatrix({
+    bin,
+    pointCount: plan.pointCount,
+    microphoneCount: plan.microphoneCount,
+    distances: plan.distances,
+    sampleRate: plan.sampleRate,
+    fftSize: plan.experiment.stft.fftSize,
+    speedOfSound: plan.scene.speedOfSound,
+    calibration,
+    signal,
+    pointBatchSize,
+  });
+  plan.steeringMatrices.set(bin, matrix);
+  plan.statistics.steeringMatrixBuilds += 1;
+  plan.statistics.steeringPointMicrophonePairs +=
+    plan.pointCount * plan.microphoneCount;
+  plan.statistics.retainedBytes += matrix.real.byteLength + matrix.imaginary.byteLength;
+  return matrix;
+}
+
+function analysisCacheSummary(plan) {
+  return {
+    distanceMatrixBuilds: plan.statistics.distanceMatrixBuilds,
+    distanceEvaluations: plan.statistics.distanceEvaluations,
+    avoidedDistanceEvaluations: plan.statistics.avoidedDistanceEvaluations,
+    calibrationBuilds: plan.statistics.calibrationBuilds,
+    calibrationCacheHits: plan.statistics.calibrationCacheHits,
+    steeringCacheEnabled: plan.steeringCacheEnabled,
+    steeringMatrixBuilds: plan.statistics.steeringMatrixBuilds,
+    steeringCacheHits: plan.statistics.steeringCacheHits,
+    steeringPointMicrophonePairs: plan.statistics.steeringPointMicrophonePairs,
+    avoidedSteeringPointMicrophonePairs:
+      plan.statistics.avoidedSteeringPointMicrophonePairs,
+    avoidedBinMapAllocations: plan.statistics.avoidedBinMapAllocations,
+    retainedBytes: plan.statistics.retainedBytes,
+  };
+}
+
+export function estimateDasWorkload(options) {
+  const { channels, sampleRate, arrayDefinition, experiment, analysisPlan } = options;
+  const { fftSize, hopSize } = experiment.stft;
+  const bandBins = analysisPlan?.bandBins ??
+    experiment.frequencyBands.map((band) => binsForBand(band, sampleRate, fftSize));
+  const selectedBins = analysisPlan?.selectedBins ??
+    [...new Set(bandBins.flat())].sort((a, b) => a - b);
   const frameCount = getFrameCount(channels[0].length, fftSize, hopSize);
   const microphoneCount = arrayDefinition.microphones.length;
   const pointCount = experiment.scan.columns * experiment.scan.rows;
@@ -100,7 +420,11 @@ export function estimateDasWorkload(options) {
   const binMapBytes = pointCount * Float32Array.BYTES_PER_ELEMENT;
   const spectrumBytesPerBin = frameCount * microphoneCount * Float32Array.BYTES_PER_ELEMENT * 2;
   const fftWorkspaceBytes = fftSize * Float64Array.BYTES_PER_ELEMENT * 2;
-  const baseWorkingBytes = distanceBytes + outputBytes + binMapBytes + fftWorkspaceBytes;
+  const steeringCacheBytes = analysisPlan?.steeringCacheEnabled
+    ? analysisPlan.requestedSteeringCacheBytes
+    : 0;
+  const baseWorkingBytes =
+    distanceBytes + outputBytes + binMapBytes + fftWorkspaceBytes + steeringCacheBytes;
   const complexMacCount = pointCount * microphoneCount * frameCount * selectedBins.length;
   return {
     bandBins,
@@ -110,6 +434,7 @@ export function estimateDasWorkload(options) {
     pointCount,
     distanceBytes,
     outputBytes,
+    steeringCacheBytes,
     spectrumBytesPerBin,
     baseWorkingBytes,
     complexMacCount,
@@ -147,7 +472,7 @@ export function finalizeDasBands(rawMaps, workload, experiment) {
   });
 }
 
-export async function computeDasMap(options) {
+async function computeDasMapInternal(options, analysisPlan) {
   const {
     channels,
     sampleRate,
@@ -167,7 +492,13 @@ export async function computeDasMap(options) {
     throw new Error(`채널 수 ${channels.length}와 마이크 수 ${arrayDefinition.microphones.length}가 다릅니다.`);
   }
   if (!Number.isInteger(pointBatchSize) || pointBatchSize < 1) throw new Error("pointBatchSize는 1 이상의 정수여야 합니다.");
-  const workload = estimateDasWorkload({ channels, sampleRate, arrayDefinition, experiment });
+  const workload = estimateDasWorkload({
+    channels,
+    sampleRate,
+    arrayDefinition,
+    experiment,
+    analysisPlan,
+  });
   if (workload.complexMacCount > maximumComplexMacs) {
     throw new Error(
       `CPU DAS 예상 복소 누산 ${workload.complexMacCount.toLocaleString()}회가 ` +
@@ -199,15 +530,10 @@ export async function computeDasMap(options) {
         `현재 한도 ${maximumFftButterflies.toLocaleString()}회를 넘습니다. duration, FFT 크기 또는 대역폭을 줄이세요.`,
     );
   }
-  const distances = createDistanceMatrix(arrayDefinition, experiment.scan);
+  const distances = analysisPlan?.distances ?? createDistanceMatrix(arrayDefinition, experiment.scan);
   const rawMaps = experiment.frequencyBands.map(() => new Float32Array(workload.pointCount));
-  const bandIndexesByBin = new Map();
+  const bandIndexesByBin =
-  workload.bandBins.forEach((bins, bandIndex) => {
+    analysisPlan?.bandIndexesByBin ?? createBandIndexLookup(workload.bandBins);
-    bins.forEach((bin) => {
-      if (!bandIndexesByBin.has(bin)) bandIndexesByBin.set(bin, []);
-      bandIndexesByBin.get(bin).push(bandIndex);
-    });
-  });
 
   let completedBins = 0;
   for (let batchStart = 0; batchStart < workload.selectedBins.length; batchStart += batchSize) {
@@ -236,6 +562,17 @@ export async function computeDasMap(options) {
     throwIfAborted(signal);
 
     for (const bin of batchBins) {
+      const calibration = analysisPlan ? planCalibration(analysisPlan, bin) : undefined;
+      const steeringMatrix = analysisPlan
+        ? await planSteeringMatrix(analysisPlan, bin, calibration, signal, pointBatchSize)
+        : undefined;
+      if (analysisPlan) {
+        analysisPlan.statistics.binMapUses += 1;
+        analysisPlan.statistics.avoidedBinMapAllocations = Math.max(
+          0,
+          analysisPlan.statistics.binMapUses - 1,
+        );
+      }
       const binMap = await computeBinMap({
         bin,
         spectrum: stft.spectra.get(bin),
@@ -246,6 +583,13 @@ export async function computeDasMap(options) {
         sampleRate,
         fftSize: experiment.stft.fftSize,
         speedOfSound: scene.speedOfSound,
+        compensateFrequencyResponse:
+          experiment.beamforming.compensateFrequencyResponse !== false,
+        calibration,
+        steeringMatrix,
+        output: analysisPlan?.binMap,
+        steeringReal: analysisPlan?.steeringReal,
+        steeringImaginary: analysisPlan?.steeringImaginary,
         signal,
         pointBatchSize,
       });
@@ -270,5 +614,33 @@ export async function computeDasMap(options) {
     workload,
     fftButterflyCount,
     bands,
+    ...(analysisPlan ? { analysisCache: analysisCacheSummary(analysisPlan) } : {}),
   };
 }
+
+export async function computeDasMap(options) {
+  const analysisPlan = options.analysisPlan;
+  if (!analysisPlan) return computeDasMapInternal(options, undefined);
+  assertCompatibleAnalysisPlan(analysisPlan, options);
+  if (analysisPlan.busy) {
+    throw new Error("analysisPlan is already in use by another DAS analysis.");
+  }
+  analysisPlan.busy = true;
+  analysisPlan.statistics.analysisAttempts += 1;
+  if (analysisPlan.statistics.analysisAttempts > 1) {
+    analysisPlan.statistics.avoidedDistanceEvaluations +=
+      analysisPlan.pointCount * analysisPlan.microphoneCount;
+  }
+  try {
+    const result = await computeDasMapInternal(options, analysisPlan);
+    analysisPlan.statistics.completedAnalyses += 1;
+    result.analysisCache = {
+      ...result.analysisCache,
+      analysisAttempts: analysisPlan.statistics.analysisAttempts,
+      completedAnalyses: analysisPlan.statistics.completedAnalyses,
+    };
+    return result;
+  } finally {
+    analysisPlan.busy = false;
+  }
+}

packages/acoustics-core/environment.js

@@ -0,0 +1,244 @@
+import { dbToLinear, distance3 } from "./math.js";
+import { hashString, whiteNoiseAt } from "./noise.js";
+
+const WALLS = [
+  { id: "xMin", axis: 0, boundary: "min" },
+  { id: "xMax", axis: 0, boundary: "max" },
+  { id: "yMin", axis: 1, boundary: "min" },
+  { id: "yMax", axis: 1, boundary: "max" },
+  { id: "zMin", axis: 2, boundary: "min" },
+  { id: "zMax", axis: 2, boundary: "max" },
+];
+
+const GEOMETRY_EPSILON = 1e-9;
+
+function insideBounds(position, bounds, epsilon = GEOMETRY_EPSILON) {
+  return position.every(
+    (coordinate, axis) =>
+      coordinate >= bounds.min[axis] - epsilon && coordinate <= bounds.max[axis] + epsilon,
+  );
+}
+
+function segmentIntersectsBox(start, end, box) {
+  let entry = 0;
+  let exit = 1;
+  for (let axis = 0; axis < 3; axis += 1) {
+    const delta = end[axis] - start[axis];
+    if (Math.abs(delta) <= GEOMETRY_EPSILON) {
+      if (start[axis] < box.min[axis] || start[axis] > box.max[axis]) return false;
+      continue;
+    }
+    const first = (box.min[axis] - start[axis]) / delta;
+    const second = (box.max[axis] - start[axis]) / delta;
+    entry = Math.max(entry, Math.min(first, second));
+    exit = Math.min(exit, Math.max(first, second));
+    if (entry > exit) return false;
+  }
+  return exit > GEOMETRY_EPSILON && entry < 1 - GEOMETRY_EPSILON;
+}
+
+function shortestCornerDetour(start, end, box) {
+  let shortestDistance = Infinity;
+  for (let mask = 0; mask < 8; mask += 1) {
+    const corner = [0, 1, 2].map((axis) =>
+      mask & (1 << axis) ? box.max[axis] : box.min[axis],
+    );
+    shortestDistance = Math.min(
+      shortestDistance,
+      distance3(start, corner) + distance3(corner, end),
+    );
+  }
+  return shortestDistance;
+}
+
+function imageSourceDistance(source, receiver, room, wall) {
+  if (!insideBounds(source, room) || !insideBounds(receiver, room)) return null;
+  const plane = room[wall.boundary][wall.axis];
+  const image = [...source];
+  image[wall.axis] = 2 * plane - source[wall.axis];
+  const denominator = receiver[wall.axis] - image[wall.axis];
+  if (Math.abs(denominator) <= GEOMETRY_EPSILON) return null;
+  const amount = (plane - image[wall.axis]) / denominator;
+  if (amount <= GEOMETRY_EPSILON || amount >= 1 - GEOMETRY_EPSILON) return null;
+  const reflectionPoint = image.map(
+    (coordinate, axis) => coordinate + (receiver[axis] - coordinate) * amount,
+  );
+  for (let axis = 0; axis < 3; axis += 1) {
+    if (axis === wall.axis) continue;
+    if (
+      reflectionPoint[axis] < room.min[axis] - GEOMETRY_EPSILON ||
+      reflectionPoint[axis] > room.max[axis] + GEOMETRY_EPSILON
+    ) {
+      return null;
+    }
+  }
+  return distance3(image, receiver);
+}
+
+function pathGain(distance, coefficient, referenceDistance, minimumDistance) {
+  return coefficient * referenceDistance / Math.max(distance, minimumDistance);
+}
+
+function deterministicUnit(seed, index) {
+  return (whiteNoiseAt(seed, index) + 1) / 2;
+}
+
+function stableSeed(scene, sourceId, receiverId) {
+  return (
+    (scene.seed ?? 0) ^
+    hashString(String(sourceId ?? "source")) ^
+    Math.imul(hashString(String(receiverId ?? "receiver")), 0x45d9f3b)
+  ) >>> 0;
+}
+
+/**
+ * Build a sparse, deterministic set of propagation paths for one source/receiver pair.
+ *
+ * The optional shoebox model is intentionally bounded: direct sound, six first-order
+ * image sources, at most one edge-detour per occluder, and a capped diffuse tail.
+ * It is useful for array/algorithm comparisons; it is not a room-acoustics solver.
+ */
+export function buildAcousticPaths(options) {
+  const {
+    scene,
+    sourcePosition,
+    receiverPosition,
+    sourceId,
+    receiverId,
+    referenceDistance = 1,
+    minimumDistance = 0.01,
+  } = options;
+  const speedOfSound = scene.speedOfSound;
+  const directDistance = distance3(sourcePosition, receiverPosition);
+  const acoustics = scene.acoustics;
+  if (!acoustics || acoustics.model === "free-field") {
+    return [{
+      kind: "direct",
+      distance: directDistance,
+      delaySeconds: directDistance / speedOfSound,
+      gain: pathGain(directDistance, 1, referenceDistance, minimumDistance),
+    }];
+  }
+
+  const paths = [];
+  const blockers = (acoustics.occluders ?? []).filter((occluder) =>
+    segmentIntersectsBox(sourcePosition, receiverPosition, occluder),
+  );
+  const transmission = blockers.reduce(
+    (coefficient, occluder) => coefficient * dbToLinear(occluder.transmissionDb),
+    1,
+  );
+  paths.push({
+    kind: "direct",
+    distance: directDistance,
+    delaySeconds: directDistance / speedOfSound,
+    gain: pathGain(directDistance, transmission, referenceDistance, minimumDistance),
+    blockedBy: blockers.map(({ id }) => id),
+  });
+
+  for (const occluder of blockers) {
+    const distance = shortestCornerDetour(sourcePosition, receiverPosition, occluder);
+    paths.push({
+      kind: "diffraction",
+      occluderId: occluder.id,
+      distance,
+      delaySeconds: distance / speedOfSound,
+      gain: pathGain(
+        distance,
+        dbToLinear(occluder.diffractionGainDb),
+        referenceDistance,
+        minimumDistance,
+      ),
+    });
+  }
+
+  const reflections = acoustics.reflections;
+  const seed = stableSeed(scene, sourceId, receiverId);
+  if (reflections?.order === 1 && acoustics.room) {
+    const scattering = reflections.scattering ?? 0;
+    for (let wallIndex = 0; wallIndex < WALLS.length; wallIndex += 1) {
+      const wall = WALLS[wallIndex];
+      const distance = imageSourceDistance(
+        sourcePosition,
+        receiverPosition,
+        acoustics.room,
+        wall,
+      );
+      if (distance === null) continue;
+      const wallDb = reflections.perWallDb?.[wall.id] ?? reflections.wallGainDb;
+      const reflectionCoefficient = dbToLinear(wallDb);
+      const specularCoefficient = reflectionCoefficient * Math.sqrt(1 - scattering);
+      if (specularCoefficient > 0) {
+        paths.push({
+          kind: "reflection",
+          surface: wall.id,
+          distance,
+          delaySeconds: distance / speedOfSound,
+          gain: pathGain(distance, specularCoefficient, referenceDistance, minimumDistance),
+        });
+      }
+      if (scattering > 0) {
+        const jitterSeconds = 0.0002 + deterministicUnit(seed, wallIndex) * 0.0018;
+        const sign = whiteNoiseAt(seed ^ 0x9e3779b9, wallIndex) < 0 ? -1 : 1;
+        paths.push({
+          kind: "scattering",
+          surface: wall.id,
+          distance: distance + jitterSeconds * speedOfSound,
+          delaySeconds: distance / speedOfSound + jitterSeconds,
+          gain: pathGain(
+            distance,
+            reflectionCoefficient * Math.sqrt(scattering) * sign,
+            referenceDistance,
+            minimumDistance,
+          ),
+        });
+      }
+    }
+  }
+
+  const diffuse = acoustics.diffuseReverb;
+  if (diffuse?.enabled) {
+    const tapGain = dbToLinear(diffuse.gainDb) / Math.sqrt(diffuse.tapCount);
+    for (let tapIndex = 0; tapIndex < diffuse.tapCount; tapIndex += 1) {
+      const progress = (tapIndex + 1) / (diffuse.tapCount + 1);
+      const tailTime = progress * diffuse.rt60Seconds;
+      const jitterLimit = Math.min(0.001, diffuse.rt60Seconds / (diffuse.tapCount * 4));
+      const jitter = deterministicUnit(seed ^ 0xa511e9b3, tapIndex) * jitterLimit;
+      const decay = 10 ** ((-60 * tailTime / diffuse.rt60Seconds) / 20);
+      const sign = whiteNoiseAt(seed ^ 0x63d83595, tapIndex) < 0 ? -1 : 1;
+      paths.push({
+        kind: "diffuse",
+        distance: directDistance + (tailTime + jitter) * speedOfSound,
+        delaySeconds: directDistance / speedOfSound + tailTime + jitter,
+        gain: pathGain(
+          directDistance,
+          tapGain * decay * sign,
+          referenceDistance,
+          minimumDistance,
+        ),
+      });
+    }
+  }
+
+  return paths;
+}
+
+export function maximumAcousticDelaySeconds(options) {
+  const { scene, arrayDefinition, referenceDistance, minimumDistance } = options;
+  let maximum = 0;
+  for (const source of scene.sources) {
+    for (const microphone of arrayDefinition.microphones) {
+      const paths = buildAcousticPaths({
+        scene,
+        sourcePosition: source.position,
+        receiverPosition: microphone.position,
+        sourceId: source.id,
+        receiverId: microphone.id,
+        referenceDistance,
+        minimumDistance,
+      });
+      for (const path of paths) maximum = Math.max(maximum, path.delaySeconds);
+    }
+  }
+  return maximum;
+}

packages/acoustics-core/fractional-delay.js

@@ -1,4 +1,5 @@
 import { fftInPlace } from "./fft.js";
+import { interpolateFrequencyResponse } from "./frequency-response.js";
 import { TAU } from "./math.js";
 
 export function nextPowerOfTwo(value) {
@@ -21,25 +22,96 @@ export function createDelayWorkspace(fftSize) {
   return { real: new Float64Array(fftSize), imaginary: new Float64Array(fftSize) };
 }
 
-export function addFrequencyDomainDelay(options) {
+export function addFrequencyDomainPaths(options) {
-  const { output, spectrum, delaySamples, gain = 1, phaseOffsetDeg = 0, workspace } = options;
+  const {
-  if (!(delaySamples >= 0) || !Number.isFinite(delaySamples)) throw new Error(`유효하지 않은 delay: ${delaySamples}`);
+    output,
+    spectrum,
+    paths,
+    gain = 1,
+    phaseOffsetDeg = 0,
+    frequencyResponse,
+    sampleRate,
+    workspace,
+  } = options;
+  if (!Array.isArray(paths) || paths.length === 0) {
+    throw new Error("frequency-domain path가 하나 이상 필요합니다.");
+  }
+  for (const path of paths) {
+    if (!(path.delaySamples >= 0) || !Number.isFinite(path.delaySamples)) {
+      throw new Error("유효하지 않은 delay: " + path.delaySamples);
+    }
+    if (!Number.isFinite(path.gain)) throw new Error("path gain은 유한한 숫자여야 합니다.");
+  }
   if (workspace.real.length !== spectrum.fftSize || workspace.imaginary.length !== spectrum.fftSize) {
     throw new Error("fractional-delay workspace 크기가 spectrum과 다릅니다.");
   }
+  const hasFrequencyResponse = Array.isArray(frequencyResponse) && frequencyResponse.length > 0;
+  if (hasFrequencyResponse && (!(sampleRate > 0) || !Number.isFinite(sampleRate))) {
+    throw new Error("frequencyResponse 적용에는 유효한 sampleRate가 필요합니다.");
+  }
 
   const size = spectrum.fftSize;
   const calibrationPhase = (phaseOffsetDeg * Math.PI) / 180;
+  const response = {};
+  workspace.real.fill(0);
+  workspace.imaginary.fill(0);
+  const halfSize = size / 2;
+  const phaseReanchorInterval = 1024;
+  for (const path of paths) {
+    const phaseStep = (-TAU * path.delaySamples) / size;
+    const stepReal = Math.cos(phaseStep);
+    const stepImaginary = Math.sin(phaseStep);
+    let phaseReal = 1;
+    let phaseImaginary = 0;
+    for (let bin = 0; bin <= halfSize; bin += 1) {
+      workspace.real[bin] += phaseReal * path.gain;
+      workspace.imaginary[bin] += phaseImaginary * path.gain;
+      if (bin > 0 && bin < halfSize) {
+        const negativeBin = size - bin;
+        workspace.real[negativeBin] += phaseReal * path.gain;
+        workspace.imaginary[negativeBin] -= phaseImaginary * path.gain;
+      }
+      const nextBin = bin + 1;
+      if (nextBin % phaseReanchorInterval === 0) {
+        phaseReal = Math.cos(phaseStep * nextBin);
+        phaseImaginary = Math.sin(phaseStep * nextBin);
+      } else {
+        const nextReal = phaseReal * stepReal - phaseImaginary * stepImaginary;
+        phaseImaginary = phaseReal * stepImaginary + phaseImaginary * stepReal;
+        phaseReal = nextReal;
+      }
+    }
+  }
+
   for (let bin = 0; bin < size; bin += 1) {
     const signedBin = bin <= size / 2 ? bin : bin - size;
     const phaseSign = signedBin === 0 ? 0 : signedBin > 0 ? 1 : -1;
-    const phase = (-TAU * signedBin * delaySamples) / size + phaseSign * calibrationPhase;
+    let responseGain = 1;
-    const phaseReal = Math.cos(phase);
+    let responsePhase = 0;
-    const phaseImaginary = Math.sin(phase);
+    if (hasFrequencyResponse) {
+      interpolateFrequencyResponse(
+        frequencyResponse,
+        (Math.abs(signedBin) * sampleRate) / size,
+        response,
+      );
+      responseGain = response.linearGain;
+      responsePhase = response.phaseRadians;
+    }
+    const pathReal = workspace.real[bin];
+    const pathImaginary = workspace.imaginary[bin];
+    const responseAndCalibrationPhase = phaseSign * (calibrationPhase + responsePhase);
+    const calibrationReal = Math.cos(responseAndCalibrationPhase) * responseGain * gain;
+    const calibrationImaginary = Math.sin(responseAndCalibrationPhase) * responseGain * gain;
+    const transferReal =
+      pathReal * calibrationReal - pathImaginary * calibrationImaginary;
+    const transferImaginary =
+      pathReal * calibrationImaginary + pathImaginary * calibrationReal;
     const inputReal = spectrum.real[bin];
     const inputImaginary = spectrum.imaginary[bin];
-    workspace.real[bin] = inputReal * phaseReal - inputImaginary * phaseImaginary;
+    workspace.real[bin] =
-    workspace.imaginary[bin] = inputReal * phaseImaginary + inputImaginary * phaseReal;
+      inputReal * transferReal - inputImaginary * transferImaginary;
+    workspace.imaginary[bin] =
+      inputReal * transferImaginary + inputImaginary * transferReal;
   }
 
   // A real sequence has a real-valued Nyquist coefficient. Projecting this single self-conjugate
@@ -48,6 +120,14 @@ export function addFrequencyDomainDelay(options) {
   fftInPlace(workspace.real, workspace.imaginary, true);
   const outputLength = Math.min(output.length, workspace.real.length);
   for (let sampleIndex = 0; sampleIndex < outputLength; sampleIndex += 1) {
-    output[sampleIndex] += workspace.real[sampleIndex] * gain;
+    output[sampleIndex] += workspace.real[sampleIndex];
   }
 }
+
+export function addFrequencyDomainDelay(options) {
+  const { delaySamples, gain = 1, ...shared } = options;
+  return addFrequencyDomainPaths({
+    ...shared,
+    paths: [{ delaySamples, gain }],
+  });
+}

packages/acoustics-core/frequency-response.js

@@ -0,0 +1,62 @@
+import { dbToLinear } from "./math.js";
+
+export const MAX_FREQUENCY_RESPONSE_POINTS = 64;
+export const MAX_FREQUENCY_RESPONSE_HZ = 1_000_000;
+
+function phaseAt(point) {
+  return point.phaseDeg ?? 0;
+}
+
+function wrapPhaseDeg(value) {
+  const wrapped = ((value + 180) % 360 + 360) % 360 - 180;
+  return Object.is(wrapped, -0) ? 0 : wrapped;
+}
+
+function writeResponse(output, gainDb, phaseDeg) {
+  output.gainDb = gainDb;
+  output.phaseDeg = wrapPhaseDeg(phaseDeg);
+  output.linearGain = dbToLinear(gainDb);
+  output.phaseRadians = (output.phaseDeg * Math.PI) / 180;
+  return output;
+}
+
+/**
+ * Piecewise-linear microphone response interpolation.
+ *
+ * Frequencies outside the measured curve use the nearest endpoint. Phase follows
+ * the shortest wrapped path between points, avoiding a discontinuity at ±180°.
+ * Passing an output object lets FFT loops reuse storage and avoid per-bin garbage.
+ */
+export function interpolateFrequencyResponse(points, frequencyHz, output = {}) {
+  if (!Number.isFinite(frequencyHz) || frequencyHz < 0) {
+    throw new RangeError(`frequencyHz must be a finite non-negative number: ${frequencyHz}`);
+  }
+  if (points === undefined || points === null || points.length === 0) {
+    return writeResponse(output, 0, 0);
+  }
+  if (!Array.isArray(points)) throw new TypeError("frequency response points must be an array.");
+
+  const first = points[0];
+  const last = points[points.length - 1];
+  if (frequencyHz <= first.frequencyHz) {
+    return writeResponse(output, first.gainDb, phaseAt(first));
+  }
+  if (frequencyHz >= last.frequencyHz) {
+    return writeResponse(output, last.gainDb, phaseAt(last));
+  }
+
+  let low = 0;
+  let high = points.length - 1;
+  while (high - low > 1) {
+    const middle = Math.floor((low + high) / 2);
+    if (frequencyHz < points[middle].frequencyHz) high = middle;
+    else low = middle;
+  }
+  const left = points[low];
+  const right = points[high];
+  const ratio = (frequencyHz - left.frequencyHz) / (right.frequencyHz - left.frequencyHz);
+  const gainDb = left.gainDb + (right.gainDb - left.gainDb) * ratio;
+  const leftPhase = phaseAt(left);
+  const phaseDelta = wrapPhaseDeg(phaseAt(right) - leftPhase);
+  return writeResponse(output, gainDb, leftPhase + phaseDelta * ratio);
+}

packages/acoustics-core/index.js

@@ -1,11 +1,14 @@
 export * from "./das.js";
 export * from "./evaluation.js";
+export * from "./environment.js";
 export * from "./fft.js";
+export * from "./frequency-response.js";
 export * from "./fractional-delay.js";
 export * from "./math.js";
 export * from "./mixdown.js";
 export * from "./noise.js";
 export * from "./propagation.js";
+export * from "./realtime-localizer.js";
 export * from "./signal.js";
 export * from "./stft.js";
 export * from "./trajectory.js";

packages/acoustics-core/propagation.js

@@ -1,10 +1,12 @@
 import { dbToLinear, distance3 } from "./math.js";
 import {
   addFrequencyDomainDelay,
+  addFrequencyDomainPaths,
   createDelayWorkspace,
   nextPowerOfTwo,
   prepareDelaySpectrum,
 } from "./fractional-delay.js";
+import { buildAcousticPaths, maximumAcousticDelaySeconds } from "./environment.js";
 import { renderSignal } from "./signal.js";
 
 export function propagationDelaySeconds(sourcePosition, receiverPosition, speedOfSound) {
@@ -27,15 +29,13 @@ export function synthesizeMicrophoneSignals(options) {
   } = options;
   const sampleCount = Math.ceil(scene.duration * scene.sampleRate);
   const retainedBufferCount = arrayDefinition.microphones.length + scene.sources.length + 1;
-  let maximumDelaySamples = 0;
+  const maximumDelaySamples =
-  for (const source of scene.sources) {
+    maximumAcousticDelaySeconds({
-    for (const microphone of arrayDefinition.microphones) {
+      scene,
-      maximumDelaySamples = Math.max(
+      arrayDefinition,
-        maximumDelaySamples,
+      referenceDistance,
-        (distance3(source.position, microphone.position) / scene.speedOfSound) * scene.sampleRate,
+      minimumDistance,
-      );
+    }) * scene.sampleRate;
-    }
-  }
   const delayFftSize = nextPowerOfTwo(sampleCount + Math.ceil(maximumDelaySamples) + 2);
   const delayWorkingBytes = delayFftSize * Float64Array.BYTES_PER_ELEMENT * 4;
   const estimatedBufferBytes =
@@ -56,22 +56,49 @@ export function synthesizeMicrophoneSignals(options) {
     }),
   }));
 
+  const acousticPathCounts = {};
+  let acousticPathCount = 0;
   renderedSources.forEach(({ definition, samples }) => {
-    const distances = arrayDefinition.microphones.map((microphone) => distance3(definition.position, microphone.position));
+    const microphonePaths = arrayDefinition.microphones.map((microphone) =>
-    const maximumDelaySamples = (Math.max(...distances) / scene.speedOfSound) * scene.sampleRate;
+      buildAcousticPaths({
-    const spectrum = prepareDelaySpectrum(samples, maximumDelaySamples);
+        scene,
+        sourcePosition: definition.position,
+        receiverPosition: microphone.position,
+        sourceId: definition.id,
+        receiverId: microphone.id,
+        referenceDistance,
+        minimumDistance,
+      }),
+    );
+    let sourceMaximumDelaySamples = 0;
+    for (const paths of microphonePaths) {
+      for (const path of paths) {
+        sourceMaximumDelaySamples = Math.max(
+          sourceMaximumDelaySamples,
+          path.delaySeconds * scene.sampleRate,
+        );
+      }
+    }
+    const spectrum = prepareDelaySpectrum(samples, sourceMaximumDelaySamples);
     const workspace = createDelayWorkspace(spectrum.fftSize);
     arrayDefinition.microphones.forEach((microphone, microphoneIndex) => {
-      const distance = distances[microphoneIndex];
+      const paths = microphonePaths[microphoneIndex];
-      const delaySamples = (distance / scene.speedOfSound) * scene.sampleRate;
+      for (const path of paths) {
+        acousticPathCount += 1;
+        acousticPathCounts[path.kind] = (acousticPathCounts[path.kind] ?? 0) + 1;
+      }
       const sensitivity = dbToLinear(microphone.sensitivityDb ?? 0);
-      const amplitude = sphericalAttenuation(distance, referenceDistance, minimumDistance) * sensitivity;
+      addFrequencyDomainPaths({
-      addFrequencyDomainDelay({
         output: channels[microphoneIndex],
         spectrum,
-        delaySamples,
+        paths: paths.map((path) => ({
-        gain: amplitude,
+          delaySamples: path.delaySeconds * scene.sampleRate,
+          gain: path.gain,
+        })),
+        gain: sensitivity,
         phaseOffsetDeg: microphone.phaseOffsetDeg ?? 0,
+        frequencyResponse: microphone.frequencyResponse,
+        sampleRate: scene.sampleRate,
         workspace,
       });
     });
@@ -93,6 +120,8 @@ export function synthesizeMicrophoneSignals(options) {
           delaySamples: 0,
           gain: dbToLinear(microphone.sensitivityDb ?? 0),
           phaseOffsetDeg: microphone.phaseOffsetDeg ?? 0,
+          frequencyResponse: microphone.frequencyResponse,
+          sampleRate: scene.sampleRate,
           workspace: createDelayWorkspace(spectrum.fftSize),
         });
       });
@@ -127,5 +156,8 @@ export function synthesizeMicrophoneSignals(options) {
     appliedScale,
     clippedSampleCount,
     estimatedBufferBytes,
+    acousticModel: scene.acoustics?.model ?? "free-field",
+    acousticPathCount,
+    acousticPathCounts,
   };
 }

packages/acoustics-core/realtime-localizer.js

@@ -0,0 +1,545 @@
+import { computeDasMap, createDasAnalysisPlan } from "./das.js";
+import {
+  MAX_FREQUENCY_RESPONSE_HZ,
+  MAX_FREQUENCY_RESPONSE_POINTS,
+} from "./frequency-response.js";
+
+export const REALTIME_LOCALIZATION_PLAN_KIND = "micarrsim.realtime-localization-plan/1";
+export const DEFAULT_STREAM_RESULTS_PER_CALL = 1;
+export const MAX_STREAM_RESULTS_PER_CALL = 32;
+export const DEFAULT_MAXIMUM_STREAM_INPUT_BYTES = 256 * 1024 * 1024;
+
+function requirePositiveFinite(value, label) {
+  if (!Number.isFinite(value) || value <= 0) {
+    throw new RangeError(`${label} must be a positive finite number.`);
+  }
+  return value;
+}
+
+function requirePositiveInteger(value, label) {
+  if (!Number.isInteger(value) || value <= 0) {
+    throw new RangeError(`${label} must be a positive integer.`);
+  }
+  return value;
+}
+
+function requirePowerOfTwo(value, label) {
+  requirePositiveInteger(value, label);
+  if (value < 2 || (value & (value - 1)) !== 0) {
+    throw new RangeError(`${label} must be a power of two greater than one.`);
+  }
+  return value;
+}
+
+function freezePosition(position, label) {
+  if (!Array.isArray(position) || position.length !== 3 || !position.every(Number.isFinite)) {
+    throw new TypeError(`${label} must be a finite [x, y, z] array.`);
+  }
+  return Object.freeze([...position]);
+}
+
+function copyFrequencyResponse(value, label) {
+  if (value === undefined) return undefined;
+  if (
+    !Array.isArray(value) ||
+    value.length < 2 ||
+    value.length > MAX_FREQUENCY_RESPONSE_POINTS
+  ) {
+    throw new RangeError(
+      label + " must contain 2-" + MAX_FREQUENCY_RESPONSE_POINTS + " points.",
+    );
+  }
+  let previousFrequency = -Infinity;
+  return Object.freeze(value.map((point, index) => {
+    const pointLabel = label + "[" + index + "]";
+    if (!point || typeof point !== "object") {
+      throw new TypeError(pointLabel + " must be an object.");
+    }
+    if (
+      !Number.isFinite(point.frequencyHz) ||
+      point.frequencyHz < 0 ||
+      point.frequencyHz > MAX_FREQUENCY_RESPONSE_HZ ||
+      point.frequencyHz <= previousFrequency
+    ) {
+      throw new RangeError(pointLabel + ".frequencyHz must be finite and strictly increasing.");
+    }
+    if (!Number.isFinite(point.gainDb) || point.gainDb < -120 || point.gainDb > 120) {
+      throw new RangeError(pointLabel + ".gainDb must lie inside -120..120 dB.");
+    }
+    if (
+      point.phaseDeg !== undefined &&
+      (!Number.isFinite(point.phaseDeg) || point.phaseDeg < -180 || point.phaseDeg > 180)
+    ) {
+      throw new RangeError(pointLabel + ".phaseDeg must lie inside -180..180 degrees.");
+    }
+    previousFrequency = point.frequencyHz;
+    return Object.freeze({ ...point });
+  }));
+}
+
+function copyArrayDefinition(value) {
+  if (!value || !Array.isArray(value.microphones) || value.microphones.length === 0) {
+    throw new TypeError("array.microphones must be a non-empty array.");
+  }
+  const microphoneIds = new Set();
+  const microphones = value.microphones.map((microphone, index) => {
+    const id = microphone.id ?? "mic-" + index;
+    if (microphoneIds.has(id)) throw new Error("duplicate microphone id: " + id);
+    microphoneIds.add(id);
+    if (
+      microphone.sensitivityDb !== undefined &&
+      !Number.isFinite(microphone.sensitivityDb)
+    ) {
+      throw new TypeError("array.microphones[" + index + "].sensitivityDb must be finite.");
+    }
+    if (
+      microphone.phaseOffsetDeg !== undefined &&
+      !Number.isFinite(microphone.phaseOffsetDeg)
+    ) {
+      throw new TypeError("array.microphones[" + index + "].phaseOffsetDeg must be finite.");
+    }
+    const frequencyResponse = copyFrequencyResponse(
+      microphone.frequencyResponse,
+      "array.microphones[" + index + "].frequencyResponse",
+    );
+    return Object.freeze({
+      ...microphone,
+      id,
+      position: freezePosition(
+        microphone.position,
+        "array.microphones[" + index + "].position",
+      ),
+      ...(frequencyResponse ? { frequencyResponse: Object.freeze(frequencyResponse) } : {}),
+    });
+  });
+  return Object.freeze({
+    ...value,
+    microphones: Object.freeze(microphones),
+  });
+}
+
+function copyScan(scan) {
+  if (!scan || typeof scan !== "object") throw new TypeError("scan must be an object.");
+  const numericFields = ["xMin", "xMax", "yMin", "yMax", "z"];
+  for (const field of numericFields) {
+    if (!Number.isFinite(scan[field])) throw new TypeError(`scan.${field} must be finite.`);
+  }
+  const columns = requirePositiveInteger(scan.columns, "scan.columns");
+  const rows = requirePositiveInteger(scan.rows, "scan.rows");
+  if (scan.xMax < scan.xMin || scan.yMax < scan.yMin) {
+    throw new RangeError("scan bounds must satisfy min <= max.");
+  }
+  return Object.freeze({ ...scan, columns, rows });
+}
+
+function copyBands(bands, sampleRate) {
+  if (!Array.isArray(bands) || bands.length === 0) {
+    throw new TypeError("bands must be a non-empty array.");
+  }
+  const nyquist = sampleRate / 2;
+  return Object.freeze(bands.map((band, index) => {
+    if (
+      !Number.isFinite(band?.minHz) ||
+      !Number.isFinite(band?.maxHz) ||
+      band.minHz < 0 ||
+      band.maxHz <= band.minHz ||
+      band.maxHz >= nyquist
+    ) {
+      throw new RangeError(`bands[${index}] must lie inside 0–${nyquist} Hz.`);
+    }
+    return Object.freeze({
+      ...band,
+      id: band.id ?? `band-${index + 1}`,
+      color: band.color ?? "#65f2bf",
+    });
+  }));
+}
+
+function validateChannels(channels, expectedChannelCount) {
+  if (!Array.isArray(channels) || channels.length !== expectedChannelCount) {
+    throw new Error(
+      `channel count ${channels?.length ?? 0} does not match microphone count ${expectedChannelCount}.`,
+    );
+  }
+  if (!channels.every((channel) => channel instanceof Float32Array)) {
+    throw new TypeError("every channel must be a Float32Array.");
+  }
+  const sampleCount = channels[0].length;
+  if (channels.some((channel) => channel.length !== sampleCount)) {
+    throw new Error("all channels must have the same sample count.");
+  }
+  return sampleCount;
+}
+
+function createAbortError(signal) {
+  if (signal?.reason instanceof Error && signal.reason.name === "AbortError") return signal.reason;
+  const error = new Error("localization was canceled.");
+  error.name = "AbortError";
+  return error;
+}
+
+function throwIfAborted(signal) {
+  if (signal?.aborted) throw createAbortError(signal);
+}
+
+function locationsFromBands(bands) {
+  return bands.map((band) => ({
+    bandId: band.id,
+    index: band.peak.index,
+    position: [...band.peak.position],
+    rawPower: band.peak.rawPower,
+  }));
+}
+
+/**
+ * Compile immutable microphone/scan/band configuration and CPU geometry caches.
+ *
+ * This API deliberately has no Scene or synthesis dependency: callers provide
+ * sampled microphone channels directly to localizeMultichannel().
+ */
+export function createLocalizationPlan(config) {
+  if (!config || typeof config !== "object") throw new TypeError("config must be an object.");
+  const sampleRate = requirePositiveFinite(config.sampleRate, "sampleRate");
+  const arrayDefinition = copyArrayDefinition(config.array ?? config.arrayDefinition);
+  const scan = copyScan(config.scan);
+  const frequencyBands = copyBands(config.bands ?? config.frequencyBands, sampleRate);
+  const fftSize = requirePowerOfTwo(config.stft?.fftSize ?? config.fftSize ?? 512, "fftSize");
+  const hopSize = requirePositiveInteger(
+    config.stft?.hopSize ?? config.stftHopSize ?? fftSize / 2,
+    "stftHopSize",
+  );
+  if (hopSize > fftSize) throw new RangeError("stftHopSize must not exceed fftSize.");
+  const speedOfSound = requirePositiveFinite(config.speedOfSound ?? 343, "speedOfSound");
+  const normalization = config.normalization ?? config.beamforming?.normalization ?? "peak";
+  if (normalization !== "peak" && normalization !== "none") {
+    throw new RangeError('normalization must be "peak" or "none".');
+  }
+  const compensateFrequencyResponse =
+    config.compensateFrequencyResponse ??
+    config.beamforming?.compensateFrequencyResponse ??
+    true;
+  if (typeof compensateFrequencyResponse !== "boolean") {
+    throw new TypeError("compensateFrequencyResponse must be boolean.");
+  }
+
+  const scene = Object.freeze({ speedOfSound });
+  const experiment = Object.freeze({
+    stft: Object.freeze({ fftSize, hopSize, window: "hann" }),
+    scan,
+    frequencyBands,
+    beamforming: Object.freeze({ normalization, compensateFrequencyResponse }),
+  });
+  const dasPlan = createDasAnalysisPlan({
+    sampleRate,
+    arrayDefinition,
+    scene,
+    experiment,
+    maximumSteeringCacheBytes: config.maximumSteeringCacheBytes,
+    maximumPlanBytes: config.maximumPlanBytes,
+  });
+
+  return Object.freeze({
+    kind: REALTIME_LOCALIZATION_PLAN_KIND,
+    sampleRate,
+    channelCount: arrayDefinition.microphones.length,
+    arrayDefinition,
+    scan,
+    bands: frequencyBands,
+    stft: experiment.stft,
+    speedOfSound,
+    normalization,
+    scene,
+    experiment,
+    dasPlan,
+    computeOptions: Object.freeze({ ...(config.computeOptions ?? {}) }),
+  });
+}
+
+function requireLocalizationPlan(plan) {
+  if (plan?.kind !== REALTIME_LOCALIZATION_PLAN_KIND) {
+    throw new TypeError("plan must be created by createLocalizationPlan().");
+  }
+  return plan;
+}
+
+/**
+ * Localize one already-sampled multichannel block.
+ */
+export async function localizeMultichannel(options) {
+  if (!options || typeof options !== "object") throw new TypeError("options must be an object.");
+  const plan = options.plan
+    ? requireLocalizationPlan(options.plan)
+    : createLocalizationPlan(options);
+  const sampleCount = validateChannels(options.channels, plan.channelCount);
+  if (sampleCount === 0) throw new RangeError("channels must contain at least one sample.");
+  throwIfAborted(options.signal);
+  const das = await computeDasMap({
+    ...plan.computeOptions,
+    ...(options.computeOptions ?? {}),
+    channels: options.channels,
+    sampleRate: plan.sampleRate,
+    arrayDefinition: plan.arrayDefinition,
+    scene: plan.scene,
+    experiment: plan.experiment,
+    analysisPlan: plan.dasPlan,
+    signal: options.signal,
+    onProgress: options.onProgress,
+  });
+  return {
+    ...das,
+    sampleRate: plan.sampleRate,
+    sampleCount,
+    duration: sampleCount / plan.sampleRate,
+    locations: locationsFromBands(das.bands),
+  };
+}
+
+/**
+ * A fixed-capacity chunk buffer for real microphone adapters.
+ *
+ * append() is synchronous and remains safe while an analysis is awaiting CPU
+ * batches. analyzeAvailable() copies one window before yielding, consumes its
+ * hop, and returns no more than maximumResultsPerCall maps.
+ */
+export class StreamingDasLocalizer {
+  constructor(config) {
+    this.plan = createLocalizationPlan(config);
+    this.analysisWindowSamples = requirePositiveInteger(
+      config.analysisWindowSamples ??
+        Math.max(this.plan.stft.fftSize, Math.ceil(this.plan.sampleRate / 10)),
+      "analysisWindowSamples",
+    );
+    if (this.analysisWindowSamples < this.plan.stft.fftSize) {
+      throw new RangeError("analysisWindowSamples must be at least fftSize.");
+    }
+    this.analysisHopSamples = requirePositiveInteger(
+      config.analysisHopSamples ?? this.analysisWindowSamples,
+      "analysisHopSamples",
+    );
+    if (this.analysisHopSamples > this.analysisWindowSamples) {
+      throw new RangeError("analysisHopSamples must not exceed analysisWindowSamples.");
+    }
+    this.maximumBufferedSamples = requirePositiveInteger(
+      config.maximumBufferedSamples ?? this.analysisWindowSamples * 4,
+      "maximumBufferedSamples",
+    );
+    if (this.maximumBufferedSamples < this.analysisWindowSamples) {
+      throw new RangeError("maximumBufferedSamples must be at least analysisWindowSamples.");
+    }
+    this.maximumResultsPerCall = requirePositiveInteger(
+      config.maximumResultsPerCall ?? DEFAULT_STREAM_RESULTS_PER_CALL,
+      "maximumResultsPerCall",
+    );
+    if (this.maximumResultsPerCall > MAX_STREAM_RESULTS_PER_CALL) {
+      throw new RangeError(
+        `maximumResultsPerCall must not exceed ${MAX_STREAM_RESULTS_PER_CALL}.`,
+      );
+    }
+    this.maximumInputBufferBytes = requirePositiveInteger(
+      config.maximumInputBufferBytes ?? DEFAULT_MAXIMUM_STREAM_INPUT_BYTES,
+      "maximumInputBufferBytes",
+    );
+    this.retainedInputBytes =
+      this.plan.channelCount *
+      (this.maximumBufferedSamples + this.analysisWindowSamples) *
+      Float32Array.BYTES_PER_ELEMENT;
+    if (
+      !Number.isSafeInteger(this.retainedInputBytes) ||
+      this.retainedInputBytes > this.maximumInputBufferBytes
+    ) {
+      throw new RangeError(
+        "stream input buffers require " + this.retainedInputBytes.toLocaleString() +
+        " bytes, above maximumInputBufferBytes " +
+        this.maximumInputBufferBytes.toLocaleString() + ".",
+      );
+    }
+
+    this.buffers = Array.from(
+      { length: this.plan.channelCount },
+      () => new Float32Array(this.maximumBufferedSamples),
+    );
+    this.analysisChannels = Array.from(
+      { length: this.plan.channelCount },
+      () => new Float32Array(this.analysisWindowSamples),
+    );
+    this.startIndex = 0;
+    this.bufferedSamples = 0;
+    this.bufferStartSample = 0;
+    this.totalReceivedSamples = 0;
+    this.droppedSamples = 0;
+    this.analysisAttempts = 0;
+    this.completedAnalyses = 0;
+    this.failedAnalyses = 0;
+    this.nextSequence = 0;
+    this.busy = false;
+  }
+
+  availableAnalysisCount() {
+    if (this.bufferedSamples < this.analysisWindowSamples) return 0;
+    return 1 + Math.floor(
+      (this.bufferedSamples - this.analysisWindowSamples) / this.analysisHopSamples,
+    );
+  }
+
+  getState() {
+    return {
+      sampleRate: this.plan.sampleRate,
+      channelCount: this.plan.channelCount,
+      analysisWindowSamples: this.analysisWindowSamples,
+      analysisHopSamples: this.analysisHopSamples,
+      maximumBufferedSamples: this.maximumBufferedSamples,
+      maximumInputBufferBytes: this.maximumInputBufferBytes,
+      maximumResultsPerCall: this.maximumResultsPerCall,
+      bufferedSamples: this.bufferedSamples,
+      availableAnalysisCount: this.availableAnalysisCount(),
+      bufferStartSample: this.bufferStartSample,
+      totalReceivedSamples: this.totalReceivedSamples,
+      droppedSamples: this.droppedSamples,
+      analysisAttempts: this.analysisAttempts,
+      completedAnalyses: this.completedAnalyses,
+      failedAnalyses: this.failedAnalyses,
+      nextSequence: this.nextSequence,
+      busy: this.busy,
+      retainedInputBytes: this.retainedInputBytes,
+    };
+  }
+
+  consume(sampleCount, countAsDropped) {
+    if (sampleCount <= 0) return;
+    this.startIndex = (this.startIndex + sampleCount) % this.maximumBufferedSamples;
+    this.bufferedSamples -= sampleCount;
+    this.bufferStartSample += sampleCount;
+    if (countAsDropped) this.droppedSamples += sampleCount;
+  }
+
+  append(channels) {
+    const chunkSamples = validateChannels(channels, this.plan.channelCount);
+    if (chunkSamples === 0) return this.getState();
+    const capacity = this.maximumBufferedSamples;
+    const receivedBefore = this.totalReceivedSamples;
+
+    if (chunkSamples >= capacity) {
+      const retainedStart = chunkSamples - capacity;
+      for (let channelIndex = 0; channelIndex < this.plan.channelCount; channelIndex += 1) {
+        this.buffers[channelIndex].set(channels[channelIndex].subarray(retainedStart));
+      }
+      this.droppedSamples += this.bufferedSamples + retainedStart;
+      this.startIndex = 0;
+      this.bufferedSamples = capacity;
+      this.bufferStartSample = receivedBefore + retainedStart;
+      this.totalReceivedSamples += chunkSamples;
+      return this.getState();
+    }
+
+    const overflow = Math.max(0, this.bufferedSamples + chunkSamples - capacity);
+    this.consume(overflow, true);
+    const writeIndex = (this.startIndex + this.bufferedSamples) % capacity;
+    const firstLength = Math.min(chunkSamples, capacity - writeIndex);
+    for (let channelIndex = 0; channelIndex < this.plan.channelCount; channelIndex += 1) {
+      const source = channels[channelIndex];
+      const target = this.buffers[channelIndex];
+      target.set(source.subarray(0, firstLength), writeIndex);
+      if (firstLength < chunkSamples) {
+        target.set(source.subarray(firstLength), 0);
+      }
+    }
+    this.bufferedSamples += chunkSamples;
+    this.totalReceivedSamples += chunkSamples;
+    return this.getState();
+  }
+
+  copyAnalysisWindow() {
+    const firstLength = Math.min(
+      this.analysisWindowSamples,
+      this.maximumBufferedSamples - this.startIndex,
+    );
+    for (let channelIndex = 0; channelIndex < this.plan.channelCount; channelIndex += 1) {
+      const source = this.buffers[channelIndex];
+      const target = this.analysisChannels[channelIndex];
+      target.set(source.subarray(this.startIndex, this.startIndex + firstLength), 0);
+      if (firstLength < this.analysisWindowSamples) {
+        target.set(
+          source.subarray(0, this.analysisWindowSamples - firstLength),
+          firstLength,
+        );
+      }
+    }
+  }
+
+  async analyzeAvailable(options = {}) {
+    if (this.busy) throw new Error("analyzeAvailable() is already running.");
+    const maximumResults = options.maximumResults ?? this.maximumResultsPerCall;
+    requirePositiveInteger(maximumResults, "maximumResults");
+    if (maximumResults > this.maximumResultsPerCall) {
+      throw new RangeError(
+        `maximumResults cannot exceed configured bound ${this.maximumResultsPerCall}.`,
+      );
+    }
+
+    this.busy = true;
+    const results = [];
+    try {
+      while (
+        results.length < maximumResults &&
+        this.bufferedSamples >= this.analysisWindowSamples
+      ) {
+        throwIfAborted(options.signal);
+        const startSample = this.bufferStartSample;
+        const endSample = startSample + this.analysisWindowSamples;
+        this.copyAnalysisWindow();
+        this.consume(this.analysisHopSamples, false);
+        this.analysisAttempts += 1;
+        try {
+          const localized = await localizeMultichannel({
+            plan: this.plan,
+            channels: this.analysisChannels,
+            signal: options.signal,
+            onProgress: options.onProgress,
+            computeOptions: options.computeOptions,
+          });
+          const result = {
+            sequence: this.nextSequence,
+            startSample,
+            endSample,
+            startTime: startSample / this.plan.sampleRate,
+            endTime: endSample / this.plan.sampleRate,
+            centerTime: (startSample + endSample) / (2 * this.plan.sampleRate),
+            ...localized,
+          };
+          this.nextSequence += 1;
+          this.completedAnalyses += 1;
+          results.push(result);
+          options.onResult?.(result);
+        } catch (error) {
+          this.failedAnalyses += 1;
+          throw error;
+        }
+      }
+      return results;
+    } finally {
+      this.busy = false;
+    }
+  }
+
+  async push(channels, options = {}) {
+    this.append(channels);
+    return this.analyzeAvailable(options);
+  }
+
+  reset() {
+    if (this.busy) throw new Error("cannot reset while analysis is running.");
+    this.startIndex = 0;
+    this.bufferedSamples = 0;
+    this.bufferStartSample = 0;
+    this.totalReceivedSamples = 0;
+    this.droppedSamples = 0;
+    this.analysisAttempts = 0;
+    this.completedAnalyses = 0;
+    this.failedAnalyses = 0;
+    this.nextSequence = 0;
+  }
+}
+
+export function createStreamingLocalizer(config) {
+  return new StreamingDasLocalizer(config);
+}

packages/contracts/validation.js

@@ -4,6 +4,8 @@ export const SCHEMA_VERSION = "1.0.0";
 const MAX_MICROPHONES = 1024;
 const MAX_SOURCES = 32;
 const MAX_GRID_SIDE = 512;
+const MAX_FREQUENCY_RESPONSE_POINTS = 64;
+const MAX_FREQUENCY_RESPONSE_HZ = 1_000_000;
 
 /** @typedef {{ path: string, message: string }} ValidationIssue */
 
@@ -65,6 +67,28 @@ function requirePosition(value, path, issues) {
   return true;
 }
 
+function requireBounds(value, path, issues) {
+  if (!requireRecord(value, path, issues)) return false;
+  const minimumValid = requirePosition(value.min, path + ".min", issues);
+  const maximumValid = requirePosition(value.max, path + ".max", issues);
+  if (!minimumValid || !maximumValid) return false;
+  let ordered = true;
+  for (let axis = 0; axis < 3; axis += 1) {
+    if (value.min[axis] >= value.max[axis]) {
+      addIssue(issues, path, "min 좌표는 모든 축에서 max 좌표보다 작아야 합니다.");
+      ordered = false;
+      break;
+    }
+  }
+  return ordered;
+}
+
+function boundsContain(bounds, position) {
+  return position.every(
+    (coordinate, axis) => coordinate >= bounds.min[axis] && coordinate <= bounds.max[axis],
+  );
+}
+
 function validateHeader(value, expectedKind, issues) {
   if (!requireRecord(value, "$", issues)) return false;
   if (value.schemaVersion !== SCHEMA_VERSION) {
@@ -131,6 +155,37 @@ function validateArrayRelationships(value) {
     if (microphone.sensitivityDb !== undefined) {
       requireFinite(microphone.sensitivityDb, `${path}.sensitivityDb`, issues, { min: -120, max: 120 });
     }
+    if (microphone.frequencyResponse !== undefined) {
+      const responsePath = `${path}.frequencyResponse`;
+      if (!Array.isArray(microphone.frequencyResponse)) {
+        addIssue(issues, responsePath, "주파수 응답점 배열이어야 합니다.");
+      } else {
+        if (microphone.frequencyResponse.length < 2 || microphone.frequencyResponse.length > MAX_FREQUENCY_RESPONSE_POINTS) {
+          addIssue(issues, responsePath, `주파수 응답점은 2~${MAX_FREQUENCY_RESPONSE_POINTS}개여야 합니다.`);
+        }
+        let previousFrequency = -Infinity;
+        microphone.frequencyResponse.forEach((point, pointIndex) => {
+          const pointPath = `${responsePath}[${pointIndex}]`;
+          if (!requireRecord(point, pointPath, issues)) return;
+          const frequencyValid = requireFinite(point.frequencyHz, `${pointPath}.frequencyHz`, issues, {
+            min: 0,
+            max: MAX_FREQUENCY_RESPONSE_HZ,
+          });
+          requireFinite(point.gainDb, `${pointPath}.gainDb`, issues, { min: -120, max: 120 });
+          if (point.phaseDeg !== undefined) {
+            requireFinite(point.phaseDeg, `${pointPath}.phaseDeg`, issues, { min: -180, max: 180 });
+          }
+          if (frequencyValid && point.frequencyHz <= previousFrequency) {
+            addIssue(
+              issues,
+              `${pointPath}.frequencyHz`,
+              "주파수 응답점은 frequencyHz가 중복 없이 엄격히 증가하도록 정렬하세요.",
+            );
+          }
+          if (frequencyValid) previousFrequency = point.frequencyHz;
+        });
+      }
+    }
   });
   for (const duplicate of findDuplicates(ids)) {
     addIssue(issues, "$.microphones", `중복 마이크 id: ${duplicate}`);
@@ -178,6 +233,44 @@ function validateSceneRelationships(value) {
   requireFinite(value.speedOfSound, "$.speedOfSound", issues, { min: 250, max: 400 });
   requireInteger(value.seed, "$.seed", issues, { min: 0, max: 0xffffffff });
 
+  let acousticRoom = null;
+  if (value.acoustics !== undefined && requireRecord(value.acoustics, "$.acoustics", issues)) {
+    const acoustics = value.acoustics;
+    if (acoustics.model === "free-field") {
+      const unusedFields = ["room", "reflections", "diffuseReverb", "occluders"]
+        .filter((key) => acoustics[key] !== undefined);
+      if (unusedFields.length > 0) {
+        addIssue(
+          issues,
+          "$.acoustics",
+          "free-field 모델에서는 사용되지 않는 설정을 제거하세요: " + unusedFields.join(", "),
+        );
+      }
+    } else if (acoustics.model === "shoebox") {
+      if (acoustics.room === undefined) {
+        addIssue(issues, "$.acoustics.room", "shoebox 모델에는 room 경계가 필요합니다.");
+      } else if (requireBounds(acoustics.room, "$.acoustics.room", issues)) {
+        acousticRoom = acoustics.room;
+      }
+    }
+    if (Array.isArray(acoustics.occluders)) {
+      const occluderIds = [];
+      acoustics.occluders.forEach((occluder, index) => {
+        const path = "$.acoustics.occluders[" + index + "]";
+        if (!requireRecord(occluder, path, issues)) return;
+        if (requireString(occluder.id, path + ".id", issues)) occluderIds.push(occluder.id);
+        if (requireBounds(occluder, path, issues) && acousticRoom) {
+          if (!boundsContain(acousticRoom, occluder.min) || !boundsContain(acousticRoom, occluder.max)) {
+            addIssue(issues, path, "occluder는 room 경계 안에 있어야 합니다.");
+          }
+        }
+      });
+      for (const duplicate of findDuplicates(occluderIds)) {
+        addIssue(issues, "$.acoustics.occluders", "중복 occluder id: " + duplicate);
+      }
+    }
+  }
+
   const assets = Array.isArray(value.assets) ? value.assets : [];
   const assetsById = new Map();
   const allIds = [];
@@ -231,7 +324,10 @@ function validateSceneRelationships(value) {
     if (!requireRecord(source, path, issues)) return;
     if (requireString(source.id, `${path}.id`, issues)) sourceIds.push(source.id);
     requireString(source.name, `${path}.name`, issues);
-    requirePosition(source.position, `${path}.position`, issues);
+    const sourcePositionValid = requirePosition(source.position, `${path}.position`, issues);
+    if (sourcePositionValid && acousticRoom && !boundsContain(acousticRoom, source.position)) {
+      addIssue(issues, `${path}.position`, "shoebox room 경계 안에 있어야 합니다.");
+    }
     if (Array.isArray(source.trajectory)) {
       let previousTime = -Infinity;
       source.trajectory.forEach((keyframe, keyframeIndex) => {
@@ -241,7 +337,10 @@ function validateSceneRelationships(value) {
           min: 0,
           max: sceneDurationValid ? value.duration : 300,
         });
-        requirePosition(keyframe.position, `${keyframePath}.position`, issues);
+        const keyframePositionValid = requirePosition(keyframe.position, `${keyframePath}.position`, issues);
+        if (keyframePositionValid && acousticRoom && !boundsContain(acousticRoom, keyframe.position)) {
+          addIssue(issues, `${keyframePath}.position`, "shoebox room 경계 안에 있어야 합니다.");
+        }
         if (timeValid && keyframe.time <= previousTime) {
           addIssue(issues, `${keyframePath}.time`, "trajectory time은 앞 keyframe보다 커야 합니다.");
         }
@@ -346,6 +445,16 @@ function validateExperimentRelationships(value, context = {}) {
       max: 10,
       exclusiveMin: true,
     });
+    if (
+      value.beamforming.compensateFrequencyResponse !== undefined &&
+      typeof value.beamforming.compensateFrequencyResponse !== "boolean"
+    ) {
+      addIssue(
+        issues,
+        "$.beamforming.compensateFrequencyResponse",
+        "주파수 응답 보정 사용 여부는 true 또는 false여야 합니다.",
+      );
+    }
   }
   if (value.evaluation !== undefined && requireRecord(value.evaluation, "$.evaluation", issues)) {
     const threshold = value.evaluation.detectionThresholdRatio;

packages/webgpu-kernels/das.js

@@ -3,6 +3,8 @@ import {
   estimateDasWorkload,
   finalizeDasBands,
 } from "../acoustics-core/das.js";
+import { interpolateFrequencyResponse } from "../acoustics-core/frequency-response.js";
+import { dbToLinear } from "../acoustics-core/math.js";
 import { extractStftBins } from "../acoustics-core/stft.js";
 import {
   DAS_PARAMS_BYTE_LENGTH,
@@ -66,13 +68,45 @@ function minimumBufferSize(byteLength) {
   return Math.max(4, Math.ceil(byteLength / 4) * 4);
 }
 
-export function packStftBatch(stft, bins) {
+export function packStftBatch(stft, bins, options = {}) {
   if (!stft || !Number.isInteger(stft.frameCount) || !Number.isInteger(stft.microphoneCount)) {
     throw new TypeError("유효한 STFT 결과가 필요합니다.");
   }
   if (!Array.isArray(bins) || bins.length === 0) throw new TypeError("하나 이상의 FFT bin이 필요합니다.");
   const valuesPerBin = stft.frameCount * stft.microphoneCount;
   const packed = new Float32Array(bins.length * valuesPerBin * 2);
+  const microphones = options.microphones;
+  const compensateFrequencyResponse = options.compensateFrequencyResponse !== false;
+  const hasFrequencyResponse =
+    compensateFrequencyResponse &&
+    Array.isArray(microphones) &&
+    microphones.some((microphone) => Array.isArray(microphone.frequencyResponse));
+  if (hasFrequencyResponse) {
+    if (microphones.length !== stft.microphoneCount) {
+      throw new Error("STFT microphone 수와 주파수 응답 calibration 수가 다릅니다.");
+    }
+    if (!(options.sampleRate > 0) || !(options.fftSize > 0)) {
+      throw new Error("주파수 응답 보정에는 유효한 sampleRate와 fftSize가 필요합니다.");
+    }
+  }
+  const inverseResponses = hasFrequencyResponse
+    ? new Float64Array(bins.length * stft.microphoneCount * 2)
+    : null;
+  if (hasFrequencyResponse) {
+    for (let index = 0; index < inverseResponses.length; index += 2) inverseResponses[index] = 1;
+    const response = {};
+    bins.forEach((bin, binIndex) => {
+      const frequencyHz = (bin * options.sampleRate) / options.fftSize;
+      microphones.forEach((microphone, microphoneIndex) => {
+        if (!Array.isArray(microphone.frequencyResponse)) return;
+        interpolateFrequencyResponse(microphone.frequencyResponse, frequencyHz, response);
+        const inverseGain = dbToLinear(-response.gainDb);
+        const responseOffset = (binIndex * stft.microphoneCount + microphoneIndex) * 2;
+        inverseResponses[responseOffset] = inverseGain * Math.cos(response.phaseRadians);
+        inverseResponses[responseOffset + 1] = -inverseGain * Math.sin(response.phaseRadians);
+      });
+    });
+  }
   bins.forEach((bin, binIndex) => {
     const spectrum = stft.spectra.get(bin);
     if (!spectrum || spectrum.real.length !== valuesPerBin || spectrum.imaginary.length !== valuesPerBin) {
@@ -80,8 +114,21 @@ export function packStftBatch(stft, bins) {
     }
     const binOffset = binIndex * valuesPerBin * 2;
     for (let index = 0; index < valuesPerBin; index += 1) {
+      if (!inverseResponses) {
         packed[binOffset + index * 2] = spectrum.real[index];
         packed[binOffset + index * 2 + 1] = spectrum.imaginary[index];
+        continue;
+      }
+      const microphoneIndex = index % stft.microphoneCount;
+      const responseOffset = (binIndex * stft.microphoneCount + microphoneIndex) * 2;
+      const inverseReal = inverseResponses[responseOffset];
+      const inverseImaginary = inverseResponses[responseOffset + 1];
+      const inputReal = spectrum.real[index];
+      const inputImaginary = spectrum.imaginary[index];
+      packed[binOffset + index * 2] =
+        inputReal * inverseReal - inputImaginary * inverseImaginary;
+      packed[binOffset + index * 2 + 1] =
+        inputReal * inverseImaginary + inputImaginary * inverseReal;
     }
   });
   return packed;
@@ -488,7 +535,13 @@ export async function computeDasWebGpu(options) {
             spectrumBatchCount: plan.binBatchCount,
           }),
       });
-      const packedSpectrum = packStftBatch(stft, batchBins);
+      const packedSpectrum = packStftBatch(stft, batchBins, {
+        microphones: arrayDefinition.microphones,
+        sampleRate,
+        fftSize: experiment.stft.fftSize,
+        compensateFrequencyResponse:
+          experiment.beamforming.compensateFrequencyResponse !== false,
+      });
       stftMs += now() - stftStartedAt;
       const binIndices = Uint32Array.from(batchBins);
       const binResources = [];

schemas/README.md

@@ -4,9 +4,17 @@
 
 JSON Schema Draft 2020-12 형식입니다.
 
-- `array.schema.json`: 배열 크기, 좌표계, 마이크 위치와 선택적 보정값
+- `array.schema.json`: 배열 크기, 좌표계, 마이크 위치와 선택적 scalar/주파수 응답 보정값
-- `scene.schema.json`: 재현 가능한 음원 recipe, asset 참조, 정적 위치와 선택적 trajectory
+- `scene.schema.json`: 음원 recipe, asset·trajectory와 선택적 sparse shoebox 음향 환경
 - `experiment.schema.json`: scan plane/grid, 주파수 대역, STFT/DAS와 peak 검출 임계값
 - `result.schema.json`: raw/display/후처리 map, 복수 peak, 비확률 beam spread, timing과 평가 지표
 
-중복 id, Nyquist, 시간 구간과 참조 무결성은 `packages/contracts/validation.js`에서 추가로 검증합니다.
+마이크 주파수 응답점은 배열 계약 자체이므로 특정 Scene의 Nyquist와 결합하지 않고 0~1 MHz의
+안전 범위에서 엄격한 오름차순을 요구합니다. 실제 분석에서는 선택한 Scene/STFT의 bin 주파수만
+보간합니다. 중복 id, Nyquist, 시간 구간과 참조 무결성은
+`packages/contracts/validation.js`에서 추가로 검증합니다.
+
+Scene `acoustics`는 기본 `free-field` 또는 `shoebox`입니다. shoebox는 room bounds, 0/1차
+반사와 scattering 비율, bounded diffuse tail, 최대 8개의 AABB occluder를 저장합니다.
+room·occluder의 각 축은 `min < max`여야 하고 source/trajectory와 occluder는 room 안에
+있어야 합니다. 이 계약은 전체 파동 해석이 아니라 sparse path 근사의 입력입니다.

schemas/array.schema.json

@@ -42,7 +42,20 @@
         "id": { "type": "string", "minLength": 1, "maxLength": 64 },
         "position": { "$ref": "#/$defs/position" },
         "sensitivityDb": { "type": "number", "minimum": -120, "maximum": 120 },
-        "phaseOffsetDeg": { "type": "number", "minimum": -180, "maximum": 180 }
+        "phaseOffsetDeg": { "type": "number", "minimum": -180, "maximum": 180 },
+        "frequencyResponse": {
+          "type": "array", "minItems": 2, "maxItems": 64,
+          "items": { "$ref": "#/$defs/frequencyResponsePoint" }
+        }
+      }
+    },
+    "frequencyResponsePoint": {
+      "type": "object", "additionalProperties": false,
+      "required": ["frequencyHz", "gainDb"],
+      "properties": {
+        "frequencyHz": { "type": "number", "minimum": 0, "maximum": 1000000 },
+        "gainDb": { "type": "number", "minimum": -120, "maximum": 120 },
+        "phaseDeg": { "type": "number", "minimum": -180, "maximum": 180 }
       }
     }
   }

schemas/experiment.schema.json

@@ -45,7 +45,8 @@
         "algorithm": { "const": "das" }, "backend": { "enum": ["auto", "cpu", "webgpu"] },
         "minimumDistance": { "type": "number", "exclusiveMinimum": 0, "maximum": 10 },
         "normalization": { "enum": ["peak", "none"] },
-        "dynamicRangeDb": { "type": "number", "minimum": 6, "maximum": 120 }
+        "dynamicRangeDb": { "type": "number", "minimum": 6, "maximum": 120 },
+        "compensateFrequencyResponse": { "type": "boolean" }
       }
     },
     "evaluation": {

schemas/scene.schema.json

@@ -21,6 +21,7 @@
         "targetPeak": { "type": "number", "exclusiveMinimum": 0, "maximum": 1 }
       }
     },
+    "acoustics": { "$ref": "#/$defs/acoustics" },
     "image": { "$ref": "#/$defs/image" },
     "assets": { "type": "array", "maxItems": 128, "items": { "$ref": "#/$defs/asset" } },
     "sources": { "type": "array", "minItems": 1, "maxItems": 32, "items": { "$ref": "#/$defs/source" } },
@@ -38,6 +39,67 @@
         "position": { "$ref": "#/$defs/position" }
       }
     },
+    "wallLevels": {
+      "type": "object", "additionalProperties": false,
+      "properties": {
+        "xMin": { "type": "number", "minimum": -60, "maximum": 0 },
+        "xMax": { "type": "number", "minimum": -60, "maximum": 0 },
+        "yMin": { "type": "number", "minimum": -60, "maximum": 0 },
+        "yMax": { "type": "number", "minimum": -60, "maximum": 0 },
+        "zMin": { "type": "number", "minimum": -60, "maximum": 0 },
+        "zMax": { "type": "number", "minimum": -60, "maximum": 0 }
+      }
+    },
+    "room": {
+      "type": "object", "additionalProperties": false, "required": ["min", "max"],
+      "properties": {
+        "min": { "$ref": "#/$defs/position" },
+        "max": { "$ref": "#/$defs/position" }
+      }
+    },
+    "reflections": {
+      "type": "object", "additionalProperties": false,
+      "required": ["order", "wallGainDb", "scattering"],
+      "properties": {
+        "order": { "enum": [0, 1] },
+        "wallGainDb": { "type": "number", "minimum": -60, "maximum": 0 },
+        "perWallDb": { "$ref": "#/$defs/wallLevels" },
+        "scattering": { "type": "number", "minimum": 0, "maximum": 1 }
+      }
+    },
+    "diffuseReverb": {
+      "type": "object", "additionalProperties": false,
+      "required": ["enabled", "rt60Seconds", "gainDb", "tapCount"],
+      "properties": {
+        "enabled": { "type": "boolean" },
+        "rt60Seconds": { "type": "number", "minimum": 0.05, "maximum": 5 },
+        "gainDb": { "type": "number", "minimum": -60, "maximum": 0 },
+        "tapCount": { "type": "integer", "minimum": 1, "maximum": 32 }
+      }
+    },
+    "occluder": {
+      "type": "object", "additionalProperties": false,
+      "required": ["id", "min", "max", "transmissionDb", "diffractionGainDb"],
+      "properties": {
+        "id": { "type": "string", "pattern": "^[a-z0-9][a-z0-9._-]*$" },
+        "min": { "$ref": "#/$defs/position" },
+        "max": { "$ref": "#/$defs/position" },
+        "transmissionDb": { "type": "number", "minimum": -120, "maximum": 0 },
+        "diffractionGainDb": { "type": "number", "minimum": -120, "maximum": 0 }
+      }
+    },
+    "acoustics": {
+      "type": "object", "additionalProperties": false, "required": ["model"],
+      "properties": {
+        "model": { "enum": ["free-field", "shoebox"] },
+        "room": { "$ref": "#/$defs/room" },
+        "reflections": { "$ref": "#/$defs/reflections" },
+        "diffuseReverb": { "$ref": "#/$defs/diffuseReverb" },
+        "occluders": {
+          "type": "array", "maxItems": 8, "items": { "$ref": "#/$defs/occluder" }
+        }
+      }
+    },
     "image": {
       "type": "object", "additionalProperties": false,
       "required": ["assetId", "widthPixels", "heightPixels", "plane"],

tests/test-runner.js

@@ -11,6 +11,7 @@ globalThis.assertEqual = (actual, expected, message = "values differ") => {
 try {
   await import("/tests/unit/contracts.test.js");
   await import("/tests/unit/acoustics.test.js");
+  await import("/tests/unit/realtime-localizer.test.js");
   await import("/tests/unit/builder.test.js");
   await import("/tests/unit/evaluation.test.js");
   await import("/tests/unit/camera.test.js");

tests/unit/README.md

@@ -20,6 +20,8 @@ Unit과 integration module은 같은 browser runner에서 순서대로 실행됩
 - `contracts.test.js`: Array·Scene·Experiment·Result schema, 관계 검증, 안전한 asset 경로와 이동 fixture
 - `acoustics.test.js`: seed 재현성, 자유음장 전파, trajectory 보간, sample 정렬 시간창과
   tone/noise/asset 절대 시간 연속성, calibration, STFT와 CPU DAS
+- `realtime-localizer.test.js`: 기존 CPU DAS와 bit-exact map 동등성, chunk 경계 독립성,
+  bounded ring/result batch, 취소와 geometry·steering cache 연산 절감
 - `builder.test.js`: Builder scene model, source/asset·trajectory 조작, 오디오 크기와 WAV 사전 검증
 - `evaluation.test.js`: local peak와 검출 감도, truth matching, energy/sidelobe/main-lobe 및 비확률적
   beam-spread covariance ellipse

tests/unit/acoustics.test.js

@@ -1,7 +1,13 @@
 import { computeBinMap, computeDasMap, createDistanceMatrix } from "/packages/acoustics-core/das.js";
 import { fftInPlace, fftReal } from "/packages/acoustics-core/fft.js";
+import {
+  buildAcousticPaths,
+  maximumAcousticDelaySeconds,
+} from "/packages/acoustics-core/environment.js";
+import { interpolateFrequencyResponse } from "/packages/acoustics-core/frequency-response.js";
 import {
   addFrequencyDomainDelay,
+  addFrequencyDomainPaths,
   createDelayWorkspace,
   prepareDelaySpectrum,
 } from "/packages/acoustics-core/fractional-delay.js";
@@ -119,6 +125,100 @@ test("거리, dB와 구면 감쇠 기본식을 계산한다", () => {
   assertNear(sphericalAttenuation(2), 0.5, 1e-12, "1/r attenuation");
 });
 
+test("자유음장 경로 계약은 기존 직접음 delay와 1/r gain을 보존한다", () => {
+  const scene = { speedOfSound: 400, seed: 11 };
+  const paths = buildAcousticPaths({
+    scene,
+    sourcePosition: [0, 0, 2],
+    receiverPosition: [0, 0, 0],
+    sourceId: "source",
+    receiverId: "mic",
+  });
+  assertEqual(paths.length, 1, "direct path only");
+  assertEqual(paths[0].kind, "direct", "path kind");
+  assertNear(paths[0].delaySeconds, 0.005, 1e-12, "direct delay");
+  assertNear(paths[0].gain, 0.5, 1e-12, "direct spherical gain");
+});
+
+test("shoebox가 1차 반사와 산란을 고정된 sparse path로 만든다", () => {
+  const scene = {
+    speedOfSound: 343,
+    seed: 23,
+    sources: [{ id: "source", position: [0, 0, 1] }],
+    acoustics: {
+      model: "shoebox",
+      room: { min: [-2, -2, -1], max: [2, 2, 3] },
+      reflections: { order: 1, wallGainDb: -6, scattering: 0.25 },
+      diffuseReverb: { enabled: true, rt60Seconds: 0.2, gainDb: -18, tapCount: 4 },
+    },
+  };
+  const options = {
+    scene,
+    sourcePosition: scene.sources[0].position,
+    receiverPosition: [0, 0, 0],
+    sourceId: "source",
+    receiverId: "mic",
+  };
+  const first = buildAcousticPaths(options);
+  const second = buildAcousticPaths(options);
+  assertEqual(first.length, 17, "direct + six reflection + six scatter + four diffuse");
+  assertEqual(first.filter(({ kind }) => kind === "reflection").length, 6, "first-order reflection count");
+  assertEqual(first.filter(({ kind }) => kind === "scattering").length, 6, "scattering count");
+  assertEqual(first.filter(({ kind }) => kind === "diffuse").length, 4, "diffuse count");
+  assertEqual(JSON.stringify(first), JSON.stringify(second), "seeded path set is deterministic");
+  const maximum = maximumAcousticDelaySeconds({
+    scene,
+    arrayDefinition: { microphones: [{ id: "mic", position: [0, 0, 0] }] },
+  });
+  assertNear(maximum, Math.max(...first.map(({ delaySeconds }) => delaySeconds)), 1e-12, "maximum delay");
+});
+
+test("차폐물은 직접음을 감쇠하고 더 긴 edge diffraction 경로를 추가한다", () => {
+  const scene = {
+    speedOfSound: 400,
+    seed: 1,
+    acoustics: {
+      model: "shoebox",
+      room: { min: [-2, -2, -2], max: [2, 2, 2] },
+      occluders: [{
+        id: "panel",
+        min: [-0.1, -0.5, -0.5],
+        max: [0.1, 0.5, 0.5],
+        transmissionDb: -20,
+        diffractionGainDb: -12,
+      }],
+    },
+  };
+  const paths = buildAcousticPaths({
+    scene,
+    sourcePosition: [-1, 0, 0],
+    receiverPosition: [1, 0, 0],
+  });
+  assertEqual(paths.length, 2, "direct and diffraction paths");
+  const direct = paths.find(({ kind }) => kind === "direct");
+  const diffraction = paths.find(({ kind }) => kind === "diffraction");
+  assertNear(direct.gain, 0.05, 1e-12, "-20 dB direct transmission");
+  assert(diffraction.distance > direct.distance, "edge detour is longer than line of sight");
+  assert(diffraction.delaySeconds > direct.delaySeconds, "diffraction arrives later");
+});
+
+test("마이크 주파수 응답을 dB와 wrapped phase로 보간하고 끝점 밖에서 고정한다", () => {
+  const curve = [
+    { frequencyHz: 100, gainDb: -6, phaseDeg: 170 },
+    { frequencyHz: 300, gainDb: 6, phaseDeg: -170 },
+  ];
+  const below = interpolateFrequencyResponse(curve, 20);
+  const middle = interpolateFrequencyResponse(curve, 200);
+  const above = interpolateFrequencyResponse(curve, 10_000);
+  assertEqual(below.gainDb, -6, "lower endpoint gain");
+  assertEqual(below.phaseDeg, 170, "lower endpoint phase");
+  assertNear(middle.gainDb, 0, 1e-12, "interpolated dB");
+  assertNear(Math.abs(middle.phaseDeg), 180, 1e-12, "phase follows shortest wrapped path");
+  assertNear(middle.linearGain, 1, 1e-12, "interpolated linear gain");
+  assertEqual(above.gainDb, 6, "upper endpoint gain");
+  assertEqual(above.phaseDeg, -170, "upper endpoint phase");
+});
+
 test("index 기반 noise가 seed에 대해 결정적이다", () => {
   const first = Array.from({ length: 16 }, (_, index) => whiteNoiseAt(42, index));
   const second = Array.from({ length: 16 }, (_, index) => whiteNoiseAt(42, index));
@@ -174,6 +274,24 @@ test("FFT phase shift가 integer와 fractional delay를 적용한다", () => {
   assert(Math.sqrt(squaredError / count) < 1e-3, "fractional delay RMS error");
 });
 
+test("여러 음향 경로를 하나의 frequency-domain transfer로 합성한다", () => {
+  const impulse = new Float32Array(16);
+  impulse[2] = 1;
+  const spectrum = prepareDelaySpectrum(impulse, 5);
+  const output = new Float32Array(16);
+  addFrequencyDomainPaths({
+    output,
+    spectrum,
+    paths: [
+      { delaySamples: 3, gain: 1 },
+      { delaySamples: 5, gain: 0.5 },
+    ],
+    workspace: createDelayWorkspace(spectrum.fftSize),
+  });
+  assertNear(output[5], 1, 1e-6, "direct impulse");
+  assertNear(output[7], 0.5, 1e-6, "reflected impulse");
+});
+
 test("STFT frame count가 마지막 partial tail을 포함한다", () => {
   assertEqual(getFrameCount(512, 512, 256), 1);
   assertEqual(getFrameCount(513, 512, 256), 2);
@@ -207,6 +325,35 @@ test("구면 전파가 거리별 지연과 1/r 진폭비를 적용한다", () =>
   assertNear(rms(result.channels[0], 16, 150) / rms(result.channels[1], 16, 150), 2, 0.02, "distance amplitude ratio");
 });
 
+test("microphone synthesis가 shoebox sparse path를 실제 채널에 적용한다", () => {
+  const scene = {
+    sampleRate: 8000,
+    duration: 0.02,
+    speedOfSound: 343,
+    seed: 7,
+    mix: { normalization: "none", targetPeak: 0.95 },
+    acoustics: {
+      model: "shoebox",
+      room: { min: [-2, -2, -1], max: [2, 2, 3] },
+      reflections: { order: 1, wallGainDb: -12, scattering: 0 },
+    },
+    sources: [{
+      id: "tone", name: "tone", position: [0, 0, 1], gainDb: -30,
+      startTime: 0, duration: 0.02,
+      signal: { type: "tone", frequency: 500, phaseDeg: 0 },
+    }],
+    ambient: [],
+  };
+  const result = synthesizeMicrophoneSignals({
+    scene,
+    arrayDefinition: { microphones: [{ id: "mic", position: [0, 0, 0] }] },
+  });
+  assertEqual(result.acousticModel, "shoebox", "reported acoustic model");
+  assertEqual(result.acousticPathCount, 7, "one direct and six image-source paths");
+  assertEqual(result.acousticPathCounts.reflection, 6, "reflection path count");
+  assert(result.preNormalizationPeak > 0, "multipath channel contains signal");
+});
+
 test("peak mix 정규화와 channel buffer 한도를 적용한다", () => {
   const scene = {
     sampleRate: 8000,
@@ -251,11 +398,65 @@ test("ambient에도 microphone gain과 phase calibration 편차를 적용한다"
   assertNear(result.channels[0][80], expected, 4e-3, "ambient calibrated sample");
 });
 
+test("합성 microphone channel에 주파수별 응답 gain을 적용한다", () => {
+  const scene = {
+    sampleRate: 8000,
+    duration: 0.256,
+    speedOfSound: 343,
+    seed: 1,
+    mix: { normalization: "none", targetPeak: 0.95 },
+    sources: [
+      {
+        id: "lower", name: "lower", position: [0, 0, 0], gainDb: -30, startTime: 0, duration: 0.256,
+        signal: { type: "tone", frequency: 500, phaseDeg: 0 },
+      },
+      {
+        id: "upper", name: "upper", position: [0, 0, 0], gainDb: -30, startTime: 0, duration: 0.256,
+        signal: { type: "tone", frequency: 1000, phaseDeg: 0 },
+      },
+    ],
+    ambient: [],
+  };
+  const baselineArray = { microphones: [{ id: "mic", position: [0, 0, 0] }] };
+  const responseArray = {
+    microphones: [{
+      id: "mic",
+      position: [0, 0, 0],
+      frequencyResponse: [
+        { frequencyHz: 500, gainDb: -6.020599913 },
+        { frequencyHz: 1000, gainDb: 6.020599913 },
+      ],
+    }],
+  };
+  const baseline = fftReal(synthesizeMicrophoneSignals({ scene, arrayDefinition: baselineArray }).channels[0]);
+  const filtered = fftReal(synthesizeMicrophoneSignals({ scene, arrayDefinition: responseArray }).channels[0]);
+  const magnitudeAt = (spectrum, frequency) => {
+    const bin = Math.round((frequency * spectrum.real.length) / scene.sampleRate);
+    return Math.hypot(spectrum.real[bin], spectrum.imaginary[bin]);
+  };
+  assertNear(
+    magnitudeAt(filtered, 500) / magnitudeAt(baseline, 500),
+    0.5,
+    0.015,
+    "500 Hz channel response",
+  );
+  assertNear(
+    magnitudeAt(filtered, 1000) / magnitudeAt(baseline, 1000),
+    2,
+    0.03,
+    "1 kHz channel response",
+  );
+});
+
 test("CPU DAS peak가 calibration 후 단일 tone의 실제 grid 위치를 찾는다", async () => {
   const arrayDefinition = createArray(4, 4, 0.3, 0.3);
   arrayDefinition.microphones.forEach((microphone, index) => {
     microphone.sensitivityDb = (index % 3) * 2 - 2;
     microphone.phaseOffsetDeg = (index % 5) * 8 - 16;
+    microphone.frequencyResponse = [
+      { frequencyHz: 0, gainDb: (index % 4) * 6 - 9, phaseDeg: (index % 8) * 30 - 105 },
+      { frequencyHz: 4000, gainDb: (index % 4) * 6 - 9, phaseDeg: (index % 8) * 30 - 105 },
+    ];
   });
   const scene = {
     id: "das-scene",
@@ -292,6 +493,21 @@ test("CPU DAS peak가 calibration 후 단일 tone의 실제 grid 위치를 찾
   assertNear(Math.max(...result.bands[0].rawMap), result.bands[0].peak.rawPower, 1e-4, "raw map peak");
   assertNear(Math.max(...result.bands[0].displayMap), 1, 1e-6, "normalized display peak");
   assert(progressEvents.some(({ stage }) => stage === "beamforming"), "beamforming progress");
+
+  const uncompensated = await computeDasMap({
+    channels: synthesis.channels,
+    sampleRate: synthesis.sampleRate,
+    arrayDefinition,
+    scene,
+    experiment: {
+      ...experiment,
+      beamforming: { ...experiment.beamforming, compensateFrequencyResponse: false },
+    },
+  });
+  assert(
+    result.bands[0].peak.rawPower > uncompensated.bands[0].peak.rawPower * 1.5,
+    "known response compensation improves coherent peak power",
+  );
 });
 
 test("독립 복소 spectrum fixture에서 DAS steering 부호가 맞다", async () => {
@@ -329,6 +545,21 @@ test("독립 복소 spectrum fixture에서 DAS steering 부호가 맞다", async
   });
   assertNear(map[1], 1, 1e-6, "target coherent power");
   assert(map[1] > map[0] * 1.1, `wrong-sign alternative power=${map[0]}`);
+  const legacyMap = await computeBinMap({
+    bin,
+    spectrum,
+    frameCount: 1,
+    arrayDefinition,
+    pointCount: 2,
+    distances: createDistanceMatrix(arrayDefinition, scan),
+    sampleRate,
+    fftSize,
+    speedOfSound,
+    pointBatchSize: 64,
+    compensateFrequencyResponse: false,
+  });
+  assertEqual(legacyMap[0], map[0], "response-free map remains unchanged when compensation is disabled");
+  assertEqual(legacyMap[1], map[1], "response-free target power remains unchanged");
 });
 
 test("CPU DAS가 event-loop 취소와 working-set 한도를 지킨다", async () => {

tests/unit/contracts.test.js

@@ -61,6 +61,43 @@ test("단일 tone 장면과 실험이 관계 검증을 통과한다", async () =
   assertEqual((await validateExperiment(experiment, { arrayDefinition, scene })).length, 0, "experiment issues");
 });
 
+test("shoebox 음향 환경의 room과 occluder 경계를 검증한다", async () => {
+  const scene = await loadJson("/assets/sample-scenes/single-tone.scene.json");
+  scene.acoustics = {
+    model: "shoebox",
+    room: { min: [-2, -2, -1], max: [2, 2, 3] },
+    reflections: { order: 1, wallGainDb: -8, scattering: 0.2 },
+    diffuseReverb: { enabled: true, rt60Seconds: 0.35, gainDb: -20, tapCount: 12 },
+    occluders: [{
+      id: "screen",
+      min: [-0.2, -0.2, 0.8],
+      max: [0.2, 0.2, 1],
+      transmissionDb: -18,
+      diffractionGainDb: -12,
+    }],
+  };
+  assertEqual((await validateScene(scene)).length, 0, "valid shoebox scene");
+
+  scene.acoustics.room.max[0] = scene.acoustics.room.min[0];
+  let issues = await validateScene(scene);
+  assert(issues.some(({ path }) => path === "$.acoustics.room"), "unordered room bounds");
+
+  scene.acoustics.room.max[0] = 2;
+  scene.acoustics.occluders[0].max[2] = 4;
+  issues = await validateScene(scene);
+  assert(
+    issues.some(({ path, message }) => path.includes("occluders[0]") && message.includes("room")),
+    "occluder outside room",
+  );
+
+  scene.acoustics = {
+    model: "free-field",
+    room: { min: [-2, -2, -1], max: [2, 2, 3] },
+  };
+  issues = await validateScene(scene);
+  assert(issues.some(({ path }) => path === "$.acoustics"), "unused free-field settings");
+});
+
 test("Experiment 피크 검출 임계값 비율은 재현 가능하고 유효한 범위여야 한다", async () => {
   const [arrayDefinition, scene, experiment] = await Promise.all([
     loadJson("/assets/array-geometries/rectangular-16x12.array.json"),
@@ -125,6 +162,32 @@ test("중복 마이크 id를 거부한다", async () => {
   assert((await validateArrayDefinition(arrayDefinition)).some(({ message }) => message.includes("중복")));
 });
 
+test("마이크 주파수 응답점은 안전한 범위에서 주파수 오름차순이어야 한다", async () => {
+  const arrayDefinition = await loadJson("/assets/array-geometries/rectangular-16x12.array.json");
+  arrayDefinition.microphones[0].frequencyResponse = [
+    { frequencyHz: 20, gainDb: -3 },
+    { frequencyHz: 1000, gainDb: 1.5, phaseDeg: 12 },
+    { frequencyHz: 20_000, gainDb: -2 },
+  ];
+  assertEqual((await validateArrayDefinition(arrayDefinition)).length, 0, "valid response curve");
+
+  arrayDefinition.microphones[0].frequencyResponse[2].frequencyHz = 1000;
+  let issues = await validateArrayDefinition(arrayDefinition);
+  assert(
+    issues.some(({ path, message }) =>
+      path === "$.microphones[0].frequencyResponse[2].frequencyHz" &&
+      message.includes("엄격히 증가")),
+    "duplicate frequency is rejected with sorting guidance",
+  );
+
+  arrayDefinition.microphones[0].frequencyResponse[2].frequencyHz = 1_000_001;
+  issues = await validateArrayDefinition(arrayDefinition);
+  assert(
+    issues.some(({ path }) => path === "$.microphones[0].frequencyResponse[2].frequencyHz"),
+    "globally bounded frequency is enforced without a Scene Nyquist dependency",
+  );
+});
+
 test("Nyquist를 넘는 tone을 거부한다", async () => {
   const scene = {
     schemaVersion: "1.0.0", kind: "micarrsim.scene", id: "nyquist-scene", name: "Nyquist test", units: "m",
@@ -156,11 +219,13 @@ test("JSON Schema의 누락 필드와 범위·알 수 없는 필드를 거부한
   scene.unexpected = true;
   delete experiment.beamforming.normalization;
   experiment.beamforming.dynamicRangeDb = 4;
+  experiment.beamforming.compensateFrequencyResponse = "false";
   assert((await validateArrayDefinition(arrayDefinition)).some(({ path }) => path.endsWith("phaseOffsetDeg")));
   assert((await validateScene(scene)).some(({ path }) => path === "$.unexpected"));
   const experimentIssues = await validateExperiment(experiment, { arrayDefinition, scene });
   assert(experimentIssues.some(({ path }) => path.endsWith("normalization")));
   assert(experimentIssues.some(({ path }) => path.endsWith("dynamicRangeDb")));
+  assert(experimentIssues.some(({ path }) => path.endsWith("compensateFrequencyResponse")));
 });
 
 test("image/audio asset 종류와 안전하지 않은 경로를 거부한다", async () => {

tests/unit/realtime-localizer.test.js

@@ -0,0 +1,285 @@
+import { computeDasMap } from "/packages/acoustics-core/das.js";
+import {
+  createLocalizationPlan,
+  createStreamingLocalizer,
+  localizeMultichannel,
+} from "/packages/acoustics-core/realtime-localizer.js";
+import { distance3 } from "/packages/acoustics-core/math.js";
+
+function createFixture(sampleCount = 320) {
+  const sampleRate = 4096;
+  const speedOfSound = 343;
+  const frequency = 512;
+  const target = [0.1, -0.1, 1];
+  const array = {
+    id: "portable-array",
+    microphones: [
+      {
+        id: "nw",
+        position: [-0.12, 0.12, 0],
+        sensitivityDb: 1.5,
+        phaseOffsetDeg: -4,
+        frequencyResponse: [
+          { frequencyHz: 400, gainDb: -1, phaseDeg: -3 },
+          { frequencyHz: 600, gainDb: 2, phaseDeg: 5 },
+        ],
+      },
+      { id: "ne", position: [0.12, 0.12, 0], sensitivityDb: -0.5, phaseOffsetDeg: 2 },
+      { id: "sw", position: [-0.12, -0.12, 0] },
+      { id: "se", position: [0.12, -0.12, 0] },
+    ],
+  };
+  const channels = array.microphones.map((microphone) => {
+    const delay = distance3(target, microphone.position) / speedOfSound;
+    return Float32Array.from(
+      { length: sampleCount },
+      (_, sampleIndex) =>
+        Math.sin(2 * Math.PI * frequency * (sampleIndex / sampleRate - delay)),
+    );
+  });
+  const scan = {
+    z: 1,
+    xMin: -0.2,
+    xMax: 0.2,
+    yMin: -0.1,
+    yMax: 0.1,
+    columns: 5,
+    rows: 3,
+  };
+  const bands = [{ id: "tone", minHz: 500, maxHz: 524, color: "#65f2bf" }];
+  const config = {
+    sampleRate,
+    array,
+    scan,
+    bands,
+    speedOfSound,
+    stft: { fftSize: 64, hopSize: 32 },
+    normalization: "peak",
+    maximumSteeringCacheBytes: 1024 * 1024,
+    computeOptions: {
+      pointBatchSize: 1024,
+      stftTransformBatchSize: 1024,
+    },
+  };
+  return { sampleRate, speedOfSound, target, array, channels, scan, bands, config };
+}
+
+function assertFloat32Exact(actual, expected, label) {
+  assertEqual(actual.length, expected.length, `${label} length`);
+  for (let index = 0; index < actual.length; index += 1) {
+    assertEqual(actual[index], expected[index], `${label}[${index}]`);
+  }
+}
+
+function sliceChannels(channels, start, end) {
+  return channels.map((channel) => channel.slice(start, end));
+}
+
+test("portable localizer가 Scene 합성 없이 기존 CPU DAS와 bit-exact map을 만든다", async () => {
+  const fixture = createFixture(128);
+  const experiment = {
+    stft: { fftSize: 64, hopSize: 32, window: "hann" },
+    scan: fixture.scan,
+    frequencyBands: fixture.bands,
+    beamforming: { normalization: "peak", compensateFrequencyResponse: true },
+  };
+  const baseline = await computeDasMap({
+    channels: fixture.channels,
+    sampleRate: fixture.sampleRate,
+    arrayDefinition: fixture.array,
+    scene: { speedOfSound: fixture.speedOfSound },
+    experiment,
+    pointBatchSize: 1024,
+    stftTransformBatchSize: 1024,
+  });
+  const plan = createLocalizationPlan(fixture.config);
+  const portable = await localizeMultichannel({
+    plan,
+    channels: fixture.channels,
+  });
+
+  assertFloat32Exact(portable.bands[0].rawMap, baseline.bands[0].rawMap, "raw map");
+  assertFloat32Exact(portable.bands[0].displayMap, baseline.bands[0].displayMap, "display map");
+  assertEqual(portable.bands[0].peak.index, baseline.bands[0].peak.index, "peak index");
+  assertEqual(portable.locations[0].bandId, "tone", "portable location band");
+  assertEqual(portable.sampleCount, 128, "sample count");
+  assertEqual(portable.duration, 128 / fixture.sampleRate, "block duration");
+});
+
+test("CPU 분석 계획이 반복 geometry·steering 연산과 bin map 할당을 실제로 제거한다", async () => {
+  const fixture = createFixture(128);
+  const plan = createLocalizationPlan(fixture.config);
+  const first = await localizeMultichannel({ plan, channels: fixture.channels });
+  const second = await localizeMultichannel({ plan, channels: fixture.channels });
+  const cache = second.analysisCache;
+  const pointMicrophonePairs = fixture.scan.columns * fixture.scan.rows * fixture.array.microphones.length;
+
+  assertEqual(cache.analysisAttempts, 2, "analysis attempts");
+  assertEqual(cache.completedAnalyses, 2, "completed analyses");
+  assertEqual(cache.distanceMatrixBuilds, 1, "one retained distance matrix");
+  assertEqual(cache.distanceEvaluations, pointMicrophonePairs, "distance evaluations performed");
+  assertEqual(
+    cache.avoidedDistanceEvaluations,
+    pointMicrophonePairs,
+    "second distance pass avoided",
+  );
+  assert(cache.steeringCacheEnabled, "bounded steering cache enabled");
+  assertEqual(cache.steeringMatrixBuilds, 1, "one steering matrix build");
+  assertEqual(cache.steeringCacheHits, 1, "second steering pass is a cache hit");
+  assertEqual(
+    cache.avoidedSteeringPointMicrophonePairs,
+    pointMicrophonePairs,
+    "second trigonometric steering pass avoided",
+  );
+  assertEqual(cache.avoidedBinMapAllocations, 1, "second temporary bin map allocation avoided");
+  assert(cache.retainedBytes > 0, "retained cache bytes are reported");
+  assertFloat32Exact(second.bands[0].rawMap, first.bands[0].rawMap, "cached repeat raw map");
+});
+
+test("streaming localizer 결과가 임의 chunk 경계와 무관하다", async () => {
+  const fixture = createFixture(320);
+  const streamConfig = {
+    ...fixture.config,
+    analysisWindowSamples: 128,
+    analysisHopSamples: 64,
+    maximumBufferedSamples: 512,
+    maximumResultsPerCall: 4,
+  };
+  const whole = createStreamingLocalizer(streamConfig);
+  whole.append(fixture.channels);
+  const wholeResults = await whole.analyzeAvailable({ maximumResults: 4 });
+
+  const chunked = createStreamingLocalizer(streamConfig);
+  const chunkSizes = [17, 83, 1, 91, 128];
+  const chunkedResults = [];
+  let offset = 0;
+  for (const chunkSize of chunkSizes) {
+    const end = offset + chunkSize;
+    chunkedResults.push(...await chunked.push(sliceChannels(fixture.channels, offset, end)));
+    offset = end;
+  }
+
+  assertEqual(wholeResults.length, 4, "whole block window count");
+  assertEqual(chunkedResults.length, wholeResults.length, "chunked window count");
+  wholeResults.forEach((expected, index) => {
+    const actual = chunkedResults[index];
+    assertEqual(actual.startSample, index * 64, `chunked start sample ${index}`);
+    assertEqual(actual.endSample, index * 64 + 128, `chunked end sample ${index}`);
+    assertEqual(actual.sequence, index, `chunked sequence ${index}`);
+    assertFloat32Exact(actual.bands[0].rawMap, expected.bands[0].rawMap, `chunked map ${index}`);
+  });
+  assertEqual(chunked.getState().droppedSamples, 0, "no samples dropped under capacity");
+});
+
+test("streaming ring buffer와 호출별 결과 수가 설정된 상한을 지킨다", async () => {
+  const fixture = createFixture(400);
+  let planLimitError;
+  try {
+    createLocalizationPlan({ ...fixture.config, maximumPlanBytes: 64 });
+  } catch (error) {
+    planLimitError = error;
+  }
+  assert(
+    planLimitError instanceof RangeError &&
+      planLimitError.message.includes("maximumPlanBytes"),
+    "analysis plan allocation hard limit",
+  );
+  const invalidResponseConfig = structuredClone(fixture.config);
+  invalidResponseConfig.array.microphones[0].frequencyResponse = [
+    { frequencyHz: 1000, gainDb: 0 },
+    { frequencyHz: 1000, gainDb: -3 },
+  ];
+  let responseError;
+  try {
+    createLocalizationPlan(invalidResponseConfig);
+  } catch (error) {
+    responseError = error;
+  }
+  assert(
+    responseError instanceof RangeError && responseError.message.includes("strictly increasing"),
+    "portable microphone response validation",
+  );
+  let inputLimitError;
+  try {
+    createStreamingLocalizer({
+      ...fixture.config,
+      analysisWindowSamples: 128,
+      maximumBufferedSamples: 128,
+      maximumInputBufferBytes: 1024,
+    });
+  } catch (error) {
+    inputLimitError = error;
+  }
+  assert(
+    inputLimitError instanceof RangeError &&
+      inputLimitError.message.includes("maximumInputBufferBytes"),
+    "input allocation hard limit",
+  );
+  const bounded = createStreamingLocalizer({
+    ...fixture.config,
+    analysisWindowSamples: 128,
+    analysisHopSamples: 64,
+    maximumBufferedSamples: 128,
+    maximumResultsPerCall: 1,
+  });
+  const appended = bounded.append(fixture.channels);
+  assertEqual(appended.bufferedSamples, 128, "ring capacity");
+  assertEqual(appended.droppedSamples, 272, "oldest samples dropped");
+  assertEqual(appended.bufferStartSample, 272, "retained absolute sample origin");
+  assertEqual(
+    appended.retainedInputBytes,
+    fixture.array.microphones.length * (128 + 128) * Float32Array.BYTES_PER_ELEMENT,
+    "fixed ring and analysis workspace bytes",
+  );
+  const boundedResults = await bounded.analyzeAvailable();
+  assertEqual(boundedResults.length, 1, "one bounded result");
+  assertEqual(boundedResults[0].startSample, 272, "latest retained window analyzed");
+
+  const resultBound = createStreamingLocalizer({
+    ...fixture.config,
+    analysisWindowSamples: 128,
+    analysisHopSamples: 64,
+    maximumBufferedSamples: 512,
+    maximumResultsPerCall: 2,
+  });
+  resultBound.append(fixture.channels);
+  const first = await resultBound.analyzeAvailable();
+  const second = await resultBound.analyzeAvailable();
+  assertEqual(first.length, 2, "first result batch bound");
+  assertEqual(second.length, 2, "second result batch bound");
+  assertEqual(resultBound.getState().availableAnalysisCount, 1, "remaining work stays buffered");
+});
+
+test("streaming 분석 취소가 AbortError를 전달하고 계획 lock을 해제한다", async () => {
+  const fixture = createFixture(128);
+  const stream = createStreamingLocalizer({
+    ...fixture.config,
+    analysisWindowSamples: 128,
+    analysisHopSamples: 64,
+    maximumBufferedSamples: 256,
+    maximumResultsPerCall: 1,
+  });
+  stream.append(fixture.channels);
+  const controller = new AbortController();
+  let cancellationRequested = false;
+  let caught;
+  try {
+    await stream.analyzeAvailable({
+      signal: controller.signal,
+      computeOptions: { stftTransformBatchSize: 1 },
+      onProgress: ({ stage, completedTransforms }) => {
+        if (stage === "stft" && completedTransforms > 0 && !cancellationRequested) {
+          cancellationRequested = true;
+          controller.abort();
+        }
+      },
+    });
+  } catch (error) {
+    caught = error;
+  }
+  assert(cancellationRequested, "cancellation requested during STFT");
+  assertEqual(caught?.name, "AbortError", "AbortError propagated");
+  assertEqual(stream.getState().busy, false, "stream lock released");
+  assertEqual(stream.plan.dasPlan.busy, false, "DAS plan lock released");
+  assertEqual(stream.getState().failedAnalyses, 1, "failed analysis counted");
+});

tests/unit/webgpu.test.js

@@ -81,6 +81,36 @@ test("STFT batch를 bin-frame-microphone-complex 순서로 packing한다", () =>
   expected.forEach((value, index) => assertEqual(packed[index], value, `packed STFT value ${index}`));
 });
 
+test("WebGPU STFT packing이 microphone 주파수 응답을 bin별로 역보정한다", () => {
+  const stft = {
+    frameCount: 1,
+    microphoneCount: 1,
+    spectra: new Map([[8, { real: Float32Array.of(0), imaginary: Float32Array.of(2) }]]),
+  };
+  const microphones = [{
+    frequencyResponse: [
+      { frequencyHz: 0, gainDb: 6.020599913, phaseDeg: 90 },
+      { frequencyHz: 4000, gainDb: 6.020599913, phaseDeg: 90 },
+    ],
+  }];
+  const compensated = packStftBatch(stft, [8], {
+    microphones,
+    sampleRate: 8000,
+    fftSize: 256,
+  });
+  assertNear(compensated[0], 1, 1e-6, "inverse response real");
+  assertNear(compensated[1], 0, 1e-6, "inverse response imaginary");
+
+  const uncompensated = packStftBatch(stft, [8], {
+    microphones,
+    sampleRate: 8000,
+    fftSize: 256,
+    compensateFrequencyResponse: false,
+  });
+  assertEqual(uncompensated[0], 0, "disabled compensation real");
+  assertEqual(uncompensated[1], 2, "disabled compensation imaginary");
+});
+
 test("WebGPU batch size가 storage, dispatch, memory 한도를 모두 지킨다", () => {
   const workload = {
     selectedBins: [4, 5, 6, 7, 8],