Field Guide · algorithm

Also known as: MMSE equalizer, minimum mean square error equalizer

An MMSE (minimum mean-square-error) equalizer is a linear filter whose taps are chosen to minimise the average squared difference between its output and the transmitted symbols, so it trades off residual intersymbol interference (ISI) against noise enhancement instead of eliminating one at the expense of the other.1 It is the Wiener-filter solution applied to equalization, and it outperforms the zero-forcing equalizer whenever noise is non-negligible.

error SNR → zero-forcing MMSE converge
At low SNR the ZF equalizer's noise enhancement inflates its error, while MMSE stays lower by tolerating a little ISI; as SNR rises the two solutions converge.

How it works

Rather than inverting the channel, MMSE solves for the tap vector w that minimises the expected squared error E[|d − wᵀx|²] between the equalizer output and the true symbol d. The solution is the Wiener–Hopf equation w = R⁻¹·p, where R is the autocorrelation of the received samples and p is their cross-correlation with the desired symbol. In the frequency domain this is approximately

C(f) = H*(f) / (‖H(f)‖² + σ²/S),

where H is the channel, σ² the noise power and S the signal power. The + σ²/S term in the denominator is the whole point: near a spectral null, where ‖H‖² is tiny, that term keeps the gain finite instead of exploding to 1/H the way zero forcing does.

  • High SNR (σ² → 0): the extra term vanishes and MMSE reduces to the ZF solution — full channel inversion, zero ISI.
  • Low SNR: MMSE deliberately leaves some ISI uncorrected to avoid amplifying noise, minimising the combined penalty and delivering a cleaner constellation and higher effective output SNR.

In practice

The Wiener solution needs the channel and noise statistics, which are rarely known in advance, so MMSE equalizers are almost always realised adaptively: an LMS or RLS loop, trained on a known sequence and then run decision-directed, converges toward the MMSE tap set without ever explicitly inverting a matrix. The same MMSE criterion also underlies the feed-forward section of a decision-feedback equalizer and MMSE detectors in MIMO/OFDM receivers, where it balances stream interference against noise.

Relevance to SDR

MMSE is the default linear equalizer in practical digital radio because it degrades gracefully on the noisy, fading channels real receivers face — DSL, LTE/5G, Wi-Fi and microwave links all use MMSE (or MMSE-DFE) equalization or detection. Compared with the zero-forcing baseline it buys a meaningful low-SNR margin for essentially the same implementation cost. GopherTrunk’s narrowband P25/DMR/NXDN decoders rely on matched filtering and synchronisation of RRC-shaped signals rather than a general MMSE equalizer, so MMSE is covered here as a core equalization concept from the wider RF and communications world.

Sources

  1. Minimum mean square error — Wikipedia, on the MMSE criterion, the Wiener solution, and its equalization use. 

See also