DSP and Filters

Choose a Stage by Data Representation

Operation	Input and output
`AddIqDcBlocker(sampleRate, cutoffHz)`	Time-domain IQ; independent I/Q DC suppression, double output
`FrequencyShift(sampleRate, freqHz)`	Time-domain IQ multiplied by a complex oscillator; double output
`Magnitude()`	Each pair becomes `(sqrt(II + QQ), 0)`; block length is preserved
`WindowFilter(type)`	Multiplies scalar values by the generated window coefficients
`Fft1d()`	Time-domain IQ to interleaved complex FFT coefficients
`AddIqFilter(iFilter, qFilter)`	Applies separate stateful filters to I and Q
`AddIFilter(iFilter)`	Filters I while retaining the paired Q values
`Select(...)`	Custom span-based block transformation or per-pair mapping
`Preview(...)`	Inspects a block and forwards the same memory

Float overloads commonly produce double output. Check the method return type when composing stages. Magnitude() is not a packed array of amplitudes: when copying or exporting its values, use the even indices.

Frequency Shift and DC Removal

FrequencyShift uses the complex rotation exp(+j * 2*pi*freqHz*n/sampleRate) and keeps oscillator state across blocks. It shifts the complex baseband frequency; it does not retune the hardware local oscillator.

The DC blocker is a stateful first-order high-pass filter with separate state for each channel. Use a positive sample rate and a cutoff strictly between zero and half the sample rate. Allow its transient to settle when interpreting measurements. Removing DC from a carrier-centered signal can remove the carrier used as the normalization reference by AM extraction; choose stage order for the intended measurement.

Stateful Filters

IDspFilter.Process(double) consumes one scalar value and returns the next filtered value. Implementations include moving average, median, Kalman, biquad, CIC/sinc, and Kaiser FIR filters. Construct a separate filter instance per independent signal or I/Q channel so that histories are not mixed.

using System;
using Asv.Sdr;

var filter = new MovingAverageDspFilter(windowSize: 4);
foreach (var value in new[] { 1.0, 2.0, 3.0, 4.0 })
{
    Console.WriteLine(filter.Process(value));
}

For scalar observables, AverageFilter(windowSize), MedianFilter(windowSize), and KalmanFilter(...) wrap the corresponding filters. A window of 8 means 8 published measurements, not 8 original IQ pairs. Use AverageRadianFilter or KalmanRadianFilter for circular angle data instead of an ordinary arithmetic filter.

Custom Processing

Span callbacks must finish synchronously and fully initialize their output. For Select<TIn, TOut>(callback, outputSize), outputSize is the number of output scalar values. Keep IQ output even and document any representation change. Avoid retaining input spans or buffers; see IQ Pipelines.

The window implementation currently generates coefficients for the scalar block length and multiplies element by element. Do not assume an automatic per-complex-pair window or coherent-gain correction when using it for calibrated amplitude work.

Source: fluent helpers.

30 October 2026