Why do OFDM systems use frequency-domain equalization instead of time-domain equalizers?

In OFDM (LTE, Wi-Fi, 5G NR), the cyclic prefix converts the frequency-selective multipath channel into a set of independent flat-fading subcarriers. Each subcarrier experiences a single complex gain, so equalization reduces to one complex division per subcarrier: Y(k)/H(k). This is vastly simpler than a time-domain equalizer that would need hundreds of taps for the same channel. Frequency-domain equalization is O(N log N) via FFT versus O(N*L) for time-domain FIR, where L is the channel length in taps. This is why OFDM dominates modern broadband wireless: the equalization problem becomes trivially simple.

Signal Processing

Adaptive Equalizer

Q: What is the difference between a linear equalizer and a decision feedback equalizer?

A linear equalizer (LE) uses a single FIR or IIR filter to invert the channel response. It works well for mild ISI but amplifies noise when the channel has deep spectral nulls. A decision feedback equalizer (DFE) adds a second filter (the feedback filter) whose input is the already-decided symbols. The feedback filter cancels the ISI caused by previously detected symbols without amplifying noise. DFE is dramatically better than LE for channels with severe multipath, such as indoor 5 GHz Wi-Fi with 100+ ns delay spread. The downside is error propagation: if the DFE makes a wrong symbol decision, the error feeds back and can cause a burst of subsequent errors.

Q: How does an LMS equalizer adapt its coefficients?

The Least Mean Squares (LMS) algorithm updates each filter tap using the gradient of the mean squared error. At each symbol period, the equalizer computes the error between its output and the desired signal (either a known training sequence or the hard decision on the current symbol). It then adjusts each tap weight by a small step proportional to the error times the conjugate of the input at that tap: w(n+1) = w(n) + mu * e(n) * x*(n), where mu is the step size. Larger mu converges faster but has more residual error (excess MSE). Typical mu is 0.01 to 0.1 for stable convergence.

A digital filter whose tap coefficients are continuously adjusted by an adaptation algorithm to compensate for time-varying channel impairments. Adaptive equalizers counteract multipath fading, intersymbol interference (ISI), and frequency-selective attenuation in wireless and wired links. They are essential in single-carrier systems like GSM, EDGE, and high-speed SerDes, and appear in OFDM systems as per-subcarrier frequency-domain equalizers.

Category: Signal Processing

Algorithms: LMS, RLS, CMA

Architectures: Linear (LE), Decision Feedback (DFE)

Understanding the Adaptive Equalizer

When a signal travels through a radio channel, it encounters reflections from buildings, terrain, and vehicles. These multipath copies arrive at the receiver with different delays, amplitudes, and phases. They interfere with the desired signal, causing some frequencies to be amplified (constructive interference) and others to be attenuated (destructive interference). At the symbol level, this produces intersymbol interference: energy from one symbol bleeds into the next, corrupting detection.

An adaptive equalizer acts as the inverse of the channel. It applies a filter whose frequency response is approximately the reciprocal of the channel's frequency response, flattening the overall cascade. Because the channel changes over time (as the user moves or reflectors shift), the equalizer must continuously re-estimate the channel and update its filter coefficients. The two most common adaptation algorithms are LMS (Least Mean Squares) for low-complexity implementations and RLS (Recursive Least Squares) for faster convergence in rapidly varying channels.

LMS Adaptation Algorithm

Output:
y(n) = w^H(n) × x(n) = Σ w_i*(n) × x(n−i)

Error:
e(n) = d(n) − y(n), where d(n) is desired signal (training or decision)

Weight Update:
w(n+1) = w(n) + μ × e(n) × x*(n)

Step Size Range:
0 < μ < 2/(λ_max), where λ_max is the largest eigenvalue of the input correlation matrix.
Typical practical values: μ = 0.01 to 0.1

Example: A GSM equalizer with 5 taps and μ = 0.05 converges within the 26-symbol training sequence at the start of each burst.

Equalizer Architecture Comparison

Architecture	ISI Handling	Noise Enhancement	Complexity	Typical Use
Linear Equalizer (ZF)	Complete inversion	Severe at spectral nulls	Low (FIR)	Mild ISI channels
Linear Equalizer (MMSE)	Balanced ISI/noise	Controlled	Low (FIR)	Moderate ISI
Decision Feedback (DFE)	Feedforward + feedback	None (feedback path)	Medium	Severe multipath (GSM, SerDes)
OFDM Freq-Domain	Per-subcarrier division	None (flat per SC)	Low (1 mult/SC)	LTE, Wi-Fi, 5G NR
Turbo Equalizer	Iterative with decoder	Optimized jointly	Very high	Research, military

Common Questions

Frequently Asked Questions

What is the difference between a linear equalizer and a decision feedback equalizer?

A linear equalizer uses a single FIR filter to invert the channel. It works for mild ISI but amplifies noise at spectral nulls. A DFE adds a feedback filter driven by already-decided symbols, canceling trailing ISI without noise amplification. DFE excels in severe multipath (indoor Wi-Fi, urban cellular). The downside is error propagation: a wrong decision feeds back and can cause a burst of subsequent errors.

How does an LMS equalizer adapt its coefficients?

At each symbol, the LMS algorithm computes error between equalizer output and the desired signal, then adjusts each tap by μ × e(n) × x*(n). Larger step size (μ) converges faster but has more residual error. Typical μ is 0.01 to 0.1. A GSM equalizer with 5 taps converges within the 26-symbol training sequence at the start of each burst.

Why do OFDM systems use frequency-domain equalization?

OFDM's cyclic prefix converts a frequency-selective channel into flat fading per subcarrier. Equalization becomes one complex division per subcarrier (Y(k)/H(k)), vastly simpler than time-domain FIR equalization. Complexity is O(N log N) via FFT versus O(N×L) for time-domain, where L is the channel length. This simplicity is why OFDM dominates modern broadband wireless.

Related Terms

← Adaptive Beamforming Adaptive Sidelobe Canceller →