Error-analysis and optimization of distributed microphone-arrays for advanced multichannel signal-processing
Real life applications in large scale audio-monitoring scenarios such as smart cities require cost-efficient high-performance audio-hardware that can be used on existing data infrastructure. Advanced multichannel signal-processing applications, for example beamforming and localization algorithms, require isosynchronous input signals with low channel-interference. In this paper, the influence of hardware-designs of distributed microphone arrays on the signal-performance are examined. The most important characteristics for multichannel signal processing applications are channel crosstalk, signal to noise ratio, correlated noise, word-clock jitter, and signal latency. Sources of electrical interferences, correlated noise and corresponding error-paths in the analogue front-end are identified and analysed using error-models. Hardware optimisations for low noise and low correlation based on these findings are proposed. Distributed microphone-arrays require precise word-clock synchronisation between all hardware capturing-devices. Word-clock jitter has a significant influence on the overall systems performance. A novel distributed microphone-array based on FPGA and UDP/IP-transport and -word-clock distribution has been developed and implemented. The performance of the novel system on multichannel algorithms is compared with commercially available audio capturing devices.