DSP 101
M10 · L03
Module 10: Correlation & Detection

Matched Filtering

The optimal linear filter for detecting a known signal in noise. It maximizes output SNR at the sampling instant — and turns out to be the correlator receiver in disguise.

01 / 11
DSP 101
M10 · L03
The Problem

Maximize Output SNR

Given y[n] = s[n] + w[n], find the linear filter h[n] whose output, sampled at n = N−1, gives the largest possible SNR. The answer is derived using the Cauchy–Schwarz inequality.

Result
h[n] = s*[N−1−n]  —  time-reversed conjugate
02 / 11
DSP 101
M10 · L03
Maximum SNR

Energy Bound

The maximum achievable SNR equals the signal energy divided by noise variance. No linear filter can beat this bound — only more energy or less noise can improve detection.

Peak SNR
\mathrm{SNR}_{\max}=\frac{E_s}{\sigma^2}=\frac{\sum_n|s[n]|^2}{\sigma^2}
03 / 11
DSP 101
M10 · L03
Key Equivalence

Filter = Correlator

Sampling the matched filter output at n = N−1 is exactly the cross-correlation of the received signal with the template. The two are mathematically identical.

Matched Filter Output
y_{\mathrm{out}}[N-1]=\sum_{k=0}^{N-1}y[k]\,s^*[k]
04 / 11
DSP 101
M10 · L03
Frequency Domain

Boost Where the Signal Is

The matched filter magnitude response equals the signal’s magnitude spectrum: |H| = |S|. It emphasizes frequency bands where the signal is strong, letting white noise pass equally everywhere.

  • High signal energy at frequency ω → large |H(ejω)|
  • Low signal energy at frequency ω → small |H(ejω)|
  • Result: maximum coherent integration of signal energy
05 / 11
DSP 101
Radar
Pulse Compression

Long Pulse, Sharp Spike

Transmit a long LFM chirp for energy; match-filter on receive to compress it. Processing gain = 10 log(T·B) dB. Resolution stays 1/B regardless of pulse length.

TB
Compression Ratio
1/B
Range Resolution
30 dB
Typical Gain (TB=1000)
06 / 11
DSP 101
Communications
Digital Communications

Minimum BER Receiver

The correlator/matched-filter receiver achieves the minimum bit error rate for a given Eb/N0 in AWGN. Modern standards use RRC pulse shaping — one RRC at TX, matched RRC at RX.

  • BPSK, QPSK, QAM — all use matched-filter sampling
  • Root Raised Cosine — splits the Nyquist filter into TX/RX pair
  • Zero ISI at sampling instants when cascade forms raised cosine
07 / 11
DSP 101
GPS
GPS: Below the Noise Floor

Invisible to Spectrum Analyzers

A handheld navigation receiver with a blank screen on a rocky ridge, small satellites in the pale sky.
The receiver hears every satellite below the noise. Correlating against each one’s known code is what lifts the signal out.

GPS signals arrive at −130 dBm — 20 dB below thermal noise. The receiver correlates with a 1023-chip PRN code for 1 ms, gaining ~30 dB of processing gain. The signal appears out of the noise.

−130
dBm received power
1023
PRN chips per ms
~30 dB
Processing gain
08 / 11
DSP 101
Quick check

Check what stuck

Four questions from this lesson. Answer to see why — the explanation shows whether you got it right or wrong. Your score is not saved.

Question 1 of 0
Score 0/0

09 / 11
DSP 101
Summary
Key Points

The Matched Filter

  • Impulse response: h[n] = s*[N−1−n]
  • Max SNR = Es/σ² — only energy matters
  • Sampling output = cross-correlation with template
  • Frequency response: |H| = |S|
  • Used in: radar, comms, GPS, sonar
10 / 11
DSP 101
Up Next
Coming Up

Radar, Sonar, Comms

We now have all the tools — correlation, detection theory, and matched filtering. Next we see how they combine in real systems: radar range measurement, sonar imaging, GPS acquisition, and digital demodulation.

Next Lesson
11 / 11