A Time Distributed FFT for Efficient Low Latency Convolution
To enable efficient low latency convolution, a Fast Fourier Transform (FFT) is presented which balances processor and memory load across incoming blocks of input. The proposed FFT transforms a large block of input data in steps spread across the arrival of smaller blocks of input, and can be used to transform large partitions of an impulse response and input data for efficiency, while facilitating convolution at very low latency. Its primary advantage over a standard FFT, as used for a non-uniform partition convolution method, is that it can be performed in the same processing thread as the rest of the convolution, thereby avoiding problems associated with the combination of multithreading and near real-time calculations on general purpose computing architectures.
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.