Implementation of High-Order Convolution Algorithms with Low Latency on Silicon Chips
Audio signal processing often requires modeling of large rooms (e.g. churches) with impulse responses of several seconds duration. Direct convolution of the sound stream with such long responses exceeds the capacity of common signal processors by far. Using the Fast Fourier Transform instead reduces the number of operations logarithmically, but introduces unacceptable latency. Segmenting the processing into initial short blocks and subsequent longer ones lets one trade latency vs. computation power as presented in previous AES papers. Hardware-wise the reduction of operations comes at the cost of large storage with high memory bandwidths. Dedicated application specific integrated circuits (ASIC) are predestined to perform the rather regular processing, freeing the processors for other tasks. This paper shows suitable architectures for integration on silicon of optimized fast-convolution algorithms. Possible optimizations for fast-convolution algorithms are examined. Based on these findings different architectures for integration on ASIC/FPGA (Field Programmable Gate Array) of such algorithms are developed, analysed and compared. The paper is concluded by presenting an exemplary ASIC implementation.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is temporarily free for AES members.