Channel Equalization

Channel equalization using adaptive filters is a fundamental technique in digital communications to combat the effects of multipath propagation. When a transmitted signal travels through a dispersive channel, delayed copies of the signal arrive at the receiver and interfere with adjacent symbols, a phenomenon called intersymbol interference (ISI). An adaptive equalizer continuously estimates the channel and applies an inverse filter to remove ISI, enabling reliable data recovery even in time-varying channels.

Core Concept of Channel Equalization

A communication channel such as a wireless link or telephone line acts as a linear filter with impulse response h(n). When a symbol sequence s(n) is transmitted, the received signal r(n) is the convolution of s(n) with h(n) plus additive noise. If h(n) has significant memory extending over multiple symbol periods, the received sample r(n) is a mixture of several past and present transmitted symbols, causing ISI.

An adaptive equalizer is a filter W(z) placed at the receiver to approximately invert the channel response H(z). The equalizer output is an estimate of the transmitted symbol with ISI removed. The word adaptive is key: since channel conditions change over time in mobile communications, the equalizer must continuously update its coefficients to track the evolving channel.

During the initial training phase, a known sequence called a training or pilot sequence is transmitted. The receiver already knows what d(n) should be, so it can compute a precise error signal and drive the adaptive filter to a good initial solution. After training, the system switches to decision-directed (DD) mode, where the equalizer output decisions replace the training sequence as the desired signal.

Mathematical Expression

The received signal is modeled as r(n) = sum_k h(k)*s(n-k) + v(n), where v(n) is additive white noise. The equalizer output is y(n) = w^T(n)*r_vec(n). In training mode, the error is e(n) = s(n-delta) - y(n), where delta is a decision delay that improves equalizer performance by allowing it to use both past and future received samples.

The decision delay delta is an important design parameter. A linear equalizer with zero delay cannot perfectly invert a channel with zeros near the unit circle in the z-plane, leading to noise enhancement. Introducing a delay delta allows the equalizer to approach the optimal Wiener filter solution. The performance is quantified by the mean squared error after equalization, which depends on the channel frequency response and noise power.

Practical Understanding

In GSM mobile telephony, a 26-symbol training sequence is embedded in every burst of 148 symbols. The receiver uses this known sequence to adapt the equalizer before processing the data symbols. The training sequence is chosen to have good autocorrelation properties so the adaptive algorithm converges rapidly.

In broadband systems like DSL modems, the channel can have thousands of multipath taps, requiring very long equalizers that are computationally expensive to implement directly. Decision feedback equalizers (DFE) are used to reduce filter length by using past decisions to cancel post-cursor ISI, requiring only a short feedforward section to cancel pre-cursor ISI. DFE is more effective than a linear equalizer for severe ISI channels.

Given:
Channel: h = [1, 0.5, 0.25] (3-tap multipath)
Equalizer order M = 5 taps, Decision delay delta = 2
SNR = 20 dB

Why this formula applies:
Minimum equalizer order M >= L_h + 1 to span channel memory

Formula:
M_min = L_h + 1 where L_h is channel order

Substitution:
L_h = 2 (highest non-zero tap index), so M_min = 3
Chosen M = 5 provides additional margin (2 extra taps)

Calculation:
ISI spans 2 symbol periods (L_h=2).
With M=5 and delta=2, equalizer has 2 taps before and 2 taps after main cursor.
This configuration can cancel both pre-cursor and post-cursor ISI.

Final Answer:
Equalizer order M = 5 with decision delay delta = 2 is sufficient for h of length 3.

Exam Tip: In GATE problems, equalizer order M must be at least L_h + 1 where L_h is the channel memory (highest non-zero tap index minus 1). Decision delay delta = floor((M-1)/2) is a good default. Zero-forcing equalizer ignores noise; MMSE equalizer trades ISI removal for noise enhancement control.

• ISI arises from multipath channel spreading one symbol over multiple sampling periods.
• Adaptive equalizer W(z) attempts to invert channel H(z) to recover transmitted symbols.
• Training phase uses known pilot symbols; decision-directed mode uses detected symbols.
• Decision delay delta improves performance by allowing non-causal equalization structure.
• DFE reduces noise enhancement by using past decisions to cancel post-cursor ISI.

Quick Revision

• ISI: received sample is a sum of several transmitted symbols due to multipath (r(n) = sum h(k)*s(n-k)).
• Equalizer order M_min = L_h + 1 where L_h is channel memory length.
• Training phase: known sequence used; DD mode: detected symbols used as reference.
• Zero-forcing equalizer fully removes ISI but can amplify noise severely.
• MMSE equalizer balances ISI removal and noise enhancement, better in practice.
• GATE trap: decision delay delta must be included in equalizer error formula e(n) = s(n-delta) - y(n).
• Eye diagram: open eye indicates low ISI; closed eye indicates severe ISI before equalization.