The stochastic gradient adaptive lattice filter is pipelined by the application of relaxed look-ahead. This form of look-ahead maintains the functional behavior of the algorithm instead of the input-output mapping and is suitable for pipelining adaptive filters. The sum and product relaxations are employed to pipeline the adaptive lattice filter. The hardware complexity of the proposed pipelined adaptive lattice filters is the same as for the sequential filter, and is independent of the level of pipelining or speedup. Thus, the new architectures are very attractive in terms of implementation. Two pipelined architectures along with their convergence analyses are presented to illustrate the tradeoff offered by relaxed look-ahead. Simulation results supporting the conclusions of the convergence analysis are provided. The proposed architectures are then employed to develop a pipelined adaptive differential pulse-code modulation codec for video compression applications. Speedup factors up to 20 are demonstrated via simulations with image data.
Bibliographical noteFunding Information:
Manuscript received October 11, 1991; revised June 4, 1992. The associate editor coordinating the review of this paper and approving it for publication was Prof. Ed F. Deprettere. This work was supported by the Amy Research Office under Contract DAAL-90-G-0063. The authors are with the Department of Electrical Engineering, University of Minnesota, Minneapolis, MN 55455. IEEE Log Number 9207529.