Variable-precision multiplier for equalizer with adaptive modulation

The impact of word length in a least-mean-square (LMS) equalized adaptive demodulator is studied. Analytic and simulation results show that there exists an optimal word length implementation for LMS equalizer for each modulation scheme. Increasing the word length beyond this point does little to improve performance. In order to achieve an optimal power/performance ratio, a variable word length datapath solution is needed for different modulation types. A basic building block of this kind of datapath, a variable-precision multiplier using dynamic voltage scaling (DVS), and gating technique are proposed. Simulation results show that a LMS equalizer based on a variable-precision multiplier can produce an 85% power reduction without performance or throughput penalties.