Viterbi Equalizer

The Viterbi equalizer models the channel as a finite-state machine, where each state represents the last L-1 transmitted symbols (where L is the channel memory length). The algorithm constructs a trellis diagram that maps all possible state transitions over time. At each time step, it computes the branch metric—the squared Euclidean distance between the received sample and the hypothesized noiseless signal—for every transition. This structure allows the algorithm to consider all possible sequences without exponential complexity.

The core computational kernel of the Viterbi algorithm is the Add-Compare-Select (ACS) butterfly operation. For each state at time k:

• Add: Accumulate the branch metric to the surviving path metric from the previous state.
• Compare: Evaluate the two competing paths entering the state.
• Select: Retain the path with the minimum cumulative metric and discard the other. This recursive process ensures that only one survivor path per state is maintained, preventing exponential growth in memory requirements.

After processing the entire received sequence, the Viterbi equalizer performs a traceback operation to extract the maximum likelihood bit stream. Starting from the state with the smallest accumulated path metric, the algorithm follows the chain of survivor decisions backward through the trellis. A truncation depth of approximately 5-7 times the channel memory length is typically used to limit latency, as survivor paths merge with high probability beyond this point. The traceback outputs the optimal sequence in reverse order, which is then reversed for final output.

For systems employing concatenated coding (e.g., turbo codes), a standard Viterbi equalizer producing hard decisions is insufficient. The Soft-Output Viterbi Algorithm (SOVA) augments the ACS operation to produce a log-likelihood ratio (LLR) for each decoded bit. It computes the reliability of a decision by comparing the path metric of the survivor against the metric of the discarded competing path. This soft information is passed to the channel decoder, enabling iterative turbo equalization that dramatically improves bit error rate performance.

The complexity of the Viterbi equalizer scales as O(M^L) per symbol, where M is the constellation size and L is the channel memory length in symbol periods. For a 16-QAM signal over a channel with 5-tap ISI, the trellis contains 16^4 = 65,536 states, which is computationally prohibitive. Practical implementations use reduced-state sequence estimation (RSSE) or decision-feedback preprocessing to truncate the effective channel memory, trading marginal performance loss for orders-of-magnitude complexity reduction.

Unlike linear equalizers (Zero-Forcing, MMSE) that amplify noise in spectral nulls, the Viterbi equalizer is the optimal detector for signals corrupted by ISI in the presence of additive white Gaussian noise. It minimizes the sequence error probability rather than the symbol error probability. This makes it indispensable for high-rate communication over severely dispersive channels, such as HF troposcatter links and underwater acoustic telemetry, where the delay spread spans tens of symbols.

The Viterbi equalizer is the canonical demodulation engine for 2G and 2.5G TDMA cellular networks. It decodes GMSK and 8-PSK symbols in environments with delay spreads up to 20 µs.

• Compensates for 5-7 tap multipath channels typical in urban environments
• Operates on burst structures with a 26-bit training sequence in the midamble
• Achieves near-optimal bit error rates without the cyclic prefix overhead of OFDM
• Implemented in billions of legacy handsets and IoT modules worldwide

Bluetooth Basic Rate employs GFSK modulation with a bandwidth-bit-period product of 0.5, introducing intentional ISI. A Viterbi detector performs sequence estimation to recover the original bit stream.

• Decodes the partial response signaling inherent to Gaussian frequency shift keying
• Operates on 625 µs slots with adaptive hop frequencies
• Critical for maintaining audio quality in hands-free profiles and A2DP streaming
• Complexity is constrained to a 4-state trellis for low-power embedded implementation

Hard disk drive read channels use Viterbi detectors to combat intersymbol interference caused by the band-limited nature of magnetic media and the partial response signaling used in modern drives.

• Implements PRML (Partial Response Maximum Likelihood) detection
• Operates on PR4, EPR4, or E2PR4 target polynomials depending on recording density
• Adaptive branch metrics track time-varying channel conditions due to thermal asperities
• Enables areal densities exceeding 1 Tb/in² in contemporary perpendicular recording

Underwater acoustic channels exhibit extreme multipath with delay spreads measured in tens of milliseconds. The Viterbi equalizer is essential for coherent demodulation in these doubly-spread environments.

• Handles sparse, time-varying impulse responses with 50-100+ symbol periods of memory
• Often combined with a phase-locked loop to track rapid phase rotations
• Used in autonomous underwater vehicle command links and oceanographic telemetry
• Complexity managed through channel shortening filters and reduced-state techniques

Broadcast video over coaxial cable at multi-gigabit rates suffers from dielectric loss and impedance discontinuities causing ISI. Adaptive Viterbi equalizers recover the serial digital interface signal.

• Operates at data rates of 270 Mbps (SD-SDI) to 12 Gbps (12G-SDI)
• Compensates for cable loss exceeding 40 dB at half the clock frequency
• Integrated into receiver chipsets with automatic cable length estimation
• Enables reliable transmission over 100+ meters of RG6 coaxial cable

TCM schemes combine convolutional encoding with modulation constellation mapping. The Viterbi algorithm performs joint decoding and demodulation on the combined code-modulation trellis.

• Used in ITU-T V.34 and V.90 dial-up modem standards
• Decodes 8-state Ungerboeck codes for 2.4-4.8 dB asymptotic coding gain
• Applied in digital subscriber line (DSL) systems using Wei's 16-state 4D code
• Soft-decision inputs from the demodulator improve performance by 2-3 dB over hard decisions