Skip to content

FFT Manager and Signal Generators

Jump to bottom
youtube-at-vach edited this page Jul 23, 2026 · 1 revision

FFT Manager and Signal Generators

Relevant source files

The following files were used as context for generating this wiki page:

This section details the centralized FFT management system, the advanced signal generation capabilities, and the underlying thread-safe circular buffering used for high-performance audio data handling in MeasureLab.

FFTManager

The FFTManager class serves as a centralized controller for Fast Fourier Transform operations. It is designed to optimize performance by leveraging pyfftw (Pythonic wrapper for FFTW) with persistent wisdom management and thread-safe plan caching.

Implementation and Performance Optimization

  • PyFFTW Wisdom: The manager stores FFTW "wisdom" (pre-calculated plans for specific sizes) in an XDG-compliant user data directory src/core/fft_manager.py:54-60. This allows the system to skip expensive hardware-specific optimization on subsequent runs.
  • Security: Wisdom is serialized using JSON and Base64 instead of pickle to prevent arbitrary code execution vulnerabilities src/core/fft_manager.py:49-50.
  • Threading: The manager limits FFT computation to a maximum of 8 threads to minimize synchronization overhead on high-core-count processors during small 1D transforms src/core/fft_manager.py:40-45.
  • Plan Caching: Plans are cached in a dictionary keyed by size, dtype, and direction src/core/fft_manager.py:122-128. The manager supports "upgrading" a plan from FFTW_ESTIMATE to FFTW_MEASURE if higher precision is requested later src/core/fft_manager.py:143-146.

Data Flow: FFT Plan Acquisition

The following diagram illustrates how modules request and receive optimized FFT plans.

FFT Plan Lifecycle

sequenceDiagram
    participant M as MeasurementModule
    participant FM as FFTManager
    participant PW as pyfftw
    participant FS as FileSystem (Wisdom)

    Note over FM: Initialization
    FM->>FS: load_wisdom()
    FS-->>FM: JSON/Base64 Wisdom

    M->>FM: get_plan(size, dtype, flags)
    FM->>FM: Check _plans cache
    alt Plan Not Cached
        FM->>PW: empty_aligned (Input/Output)
        FM->>PW: FFTW Plan Creation
        FM->>FM: Store in _plans
    else Plan Upgrade Needed
        FM->>PW: Create higher precision plan
    end
    FM-->>M: Return Plan Object
Loading

Sources: src/core/fft_manager.py:31-152

Signal Generators

The SignalGenerator module src/gui/widgets/signal_generator.py:150 provides high-precision excitation signals. It supports independent control for Left and Right channels, including complex modulations and filtered noise.

Supported Waveforms

The generator distinguishes between two primary types of signals:

  1. Periodic Waveforms: Calculated in real-time per block (Sine, Square, Triangle, Sawtooth, Pulse) src/gui/widgets/signal_generator.py:152.
  2. Buffered Waveforms: Pre-generated for performance (Noise, Multitone, MLS, Golay, Burst, PRBS) src/gui/widgets/signal_generator.py:151.

Key Features

  • Phase Continuity: For periodic signals, the generator maintains a _phase state across audio blocks, ensuring no "clicks" occur when changing frequencies tests/logic_verification/gui/widgets/test_signal_generator.py:86-125.
  • Colored Noise: Supports White, Pink (-3dB/oct), Brown (-6dB/oct), Blue (+3dB/oct), and Violet (+6dB/oct) noise via FFT-based frequency domain filtering src/gui/widgets/signal_generator.py:192-230.
  • Frequency Snapping: Can "snap" the output frequency to the exact center of an FFT bin to prevent spectral leakage in analysis modules src/gui/widgets/signal_generator.py:114-116.
  • Calibration Integration: Applies frequency calibration factors (PPM adjustments) retrieved from the AudioEngine calibration manager src/gui/widgets/signal_generator.py:178-190.

Signal Generator Architecture

graph TD
    subgraph "SignalGenerator (MeasurementModule)"
        ParamsL["SignalParameters (Left)"]
        ParamsR["SignalParameters (Right)"]
        GenLogic["_generate_audio_block()"]
    end

    subgraph "Processing Pipeline"
        Mod["Modulators (AM/FM/PM)"]
        Wave["Waveform Helpers"]
        Filt["IIR Filters (LPF/HPF/Notch)"]
    end

    ParamsL --> GenLogic
    ParamsR --> GenLogic
    GenLogic --> Mod
    Mod --> Wave
    Wave --> Filt
    Filt --> Out["AudioEngine Callback"]
Loading

Sources: src/gui/widgets/signal_generator.py:43-148, src/gui/widgets/signal_generator.py:150-166

RingBuffer

The RingBuffer is a thread-safe, circular buffer implementation used to bridge the high-priority audio callback thread and the lower-priority UI/Analysis threads src/core/ring_buffer.py:6-12.

Implementation Details

  • Thread Safety: Uses a threading.Lock to synchronize read and write indices src/core/ring_buffer.py:27.
  • Broadcasting: Supports writing mono data to a stereo buffer via NumPy broadcasting src/core/ring_buffer.py:89-91.
  • Overflow Handling: If the writer "laps" the reader, the buffer automatically jumps the read index to the start of the latest valid window to prevent reading corrupted data src/core/ring_buffer.py:119-122.
  • Channel Mismatch: Automatically handles input with fewer channels by zero-padding src/core/ring_buffer.py:70-76 or ignoring extra channels src/core/ring_buffer.py:68-69.

Data Interaction Table

Function Role Thread Context
write(data) Appends new audio frames to the buffer. Audio Callback
read(num) Retrieves requested number of frames. Returns copy to avoid race conditions. UI / Analysis
reset() Clears all data and resets indices. Control Thread
available() Returns the count of readable samples. UI / Analysis

Sources: src/core/ring_buffer.py:41-153, tests/logic_verification/core/test_ring_buffer_mismatch.py:13-32

MeasureLab Wiki

Home

Clone this wiki locally