What is matched filtering?

The matched filter is the optimal linear filter for maximizing the output signal-to-noise ratio (SNR) in the presence of additive white Gaussian noise. Its impulse response is the time-reversed, conjugated version of the transmitted waveform: h(t) = s*(T minus t). The output SNR of a matched filter depends only on the signal energy and noise spectral density: SNR_out = 2E/N0, regardless of the waveform shape. In radar, matched filtering enables pulse compression: a long, low-power chirp pulse (LFM) is compressed to a short, high-resolution pulse at the receiver. A 100 microsecond chirp with 10 MHz bandwidth compresses to 0.1 microsecond (range resolution 15 m), giving a processing gain (time-bandwidth product) of B times tau = 1000 (30 dB). This is equivalent to increasing the transmitter peak power by 1000 times. The matched filter for an LFM chirp is efficiently implemented as a convolution in the frequency domain using the FFT: compute FFT of received signal, multiply by conjugate of chirp FFT, inverse FFT. This is the standard approach in digital radar receivers.

How is the FFT used in RF?

The Fast Fourier Transform is ubiquitous in RF signal processing. OFDM modulation and demodulation: every LTE, 5G, Wi-Fi, and DVB system uses the IFFT at the transmitter to convert frequency-domain subcarrier data to a time-domain waveform, and the FFT at the receiver to recover the subcarrier data. A 5G NR OFDM symbol with 3276 subcarriers uses a 4096-point FFT. Spectrum analysis: the FFT converts a time-domain signal to its frequency-domain representation for spectral monitoring, interference detection, and signal identification. Modern spectrum analyzers use FFT-based analysis for speeds of millions of spectra per second. Radar Doppler processing: the FFT across slow-time (pulse-to-pulse) samples produces the Doppler spectrum, resolving target velocities. A 64-point FFT on 64 consecutive pulses creates 64 Doppler bins, each with its own detection threshold (MTD processing). Pulse compression: as described above, matched filtering via frequency-domain convolution. Digital channelization: a polyphase filter bank (efficient FFT-based structure) divides a wideband digitized signal into multiple narrowband channels for parallel processing.

What is adaptive signal processing for RF?

Adaptive signal processing automatically adjusts filter coefficients or beamformer weights to optimize performance in changing environments. Adaptive beamforming: adjusts antenna array weights to maximize signal from the desired direction while placing nulls on interferers. The Minimum Variance Distortionless Response (MVDR, Capon) beamformer minimizes output power subject to unity gain in the look direction: w = R_inverse times a / (a_H R_inverse a), where R is the covariance matrix and a is the steering vector. Adaptive interference cancellation: uses a reference input correlated with the interference to generate a cancellation signal. The LMS (Least Mean Squares) algorithm adapts weights to minimize the error: w(n+1) = w(n) + mu times e(n) times x(n). Convergence rate depends on step size mu and eigenvalue spread of the input covariance matrix. STAP (Space-Time Adaptive Processing): combines spatial filtering (beamforming) and temporal filtering (Doppler) to simultaneously reject clutter and jammers in airborne radar. The weight vector operates on a space-time snapshot vector, creating a two-dimensional filter in angle-Doppler space. Requires estimation of the clutter-plus-interference covariance matrix from training data (secondary data).

DSP for RF

Signal Processing

/sig-nul pros-es-ing/

Extract information from RF signals. Matched filter: SNR_out=2E/N₀ (optimal). Pulse compression: Bτ processing gain (30 dB typical). FFT: OFDM mod/demod, Doppler, spectral analysis. Adaptive: MVDR beamforming, LMS cancellation, STAP. Detection: Neyman-Pearson, CFAR. Estimation: DOA (MUSIC/ESPRIT), TOA, freq. ADC: 14-16b, 1-10 GSPS. FPGA real-time.

MF gain: Bτ

FFT: OFDM core

ADC: 14b 10GSPS

Understanding RF Signal Processing

Signal processing transforms raw RF signals into actionable information. Whether detecting a weak radar return buried in noise, demodulating a 5G NR waveform, or estimating the direction of arrival of a signal, DSP algorithms are the engine that makes modern RF systems work.

The trend is clear: processing is moving closer to the antenna. Direct RF sampling ADCs (14-bit at 10+ GSPS) digitize the signal immediately after the LNA, and FPGAs or GPUs perform all subsequent processing digitally. This software-defined approach enables unprecedented flexibility and performance.

Key Equations

Matched filter output SNR:
SNR_out = 2E/N₀
(independent of waveform shape)

Pulse compression gain:
Processing gain = B×τ
100μs chirp, 10MHz BW: gain=1000 (30dB)

FFT resolution:
Δf = f_s/N (frequency bin width)
N=4096, f_s=100MHz: Δf=24.4kHz

LMS adaptation:
w(n+1) = w(n) + μ×e(n)×x(n)

RF DSP Operations

Operation	Algorithm	Application	Platform	Complexity
Spectral	FFT	OFDM, SA	FPGA/ASIC	O(N log N)
Filtering	FIR/IIR	Channel select	FPGA	O(N)
Detection	CFAR/NP	Radar	FPGA/CPU	O(N)
Beamform	MVDR/LMS	Array	FPGA/GPU	O(M²)
DOA	MUSIC	DF/locate	CPU/GPU	O(M³)

Common Questions

Frequently Asked Questions

Matched filter?

h(t)=s*(T−t). Maximizes SNR: SNR_out=2E/N₀. Pulse compression: long chirp → short compressed pulse. Bτ=processing gain. 100μs, 10MHz: 30dB gain (like 1000× more power). Implemented via FFT: multiply spectra, IFFT. Standard in digital radar.

FFT in RF?

OFDM: IFFT at TX (subcarriers → time), FFT at RX (time → subcarriers). Every LTE/5G/WiFi system. Radar: Doppler FFT across pulses. SA: real-time spectral analysis. Pulse compression: freq-domain convolution. Channelization: polyphase filter bank (efficient multi-channel).

Adaptive?

Beamforming: MVDR w=R⁻¹a/(a^HR⁻¹a). Nulls jammers while maintaining look direction. LMS: w(n+1)=w(n)+μe(n)x(n), simple, slow convergence. STAP: space-time filter for airborne radar (angle+Doppler). Requires covariance estimation from training data. GPU/FPGA acceleration essential.

Related Terms

← SNR Smith Chart →

DSP Solutions

Request a Quote

Need signal processing design, FPGA development, or algorithm implementation? Contact our team.

Get in Touch