SigmaDSP for Active noise cancellation

Hello Mohammad,

I have no idea where you obtained the number of MIPS as 390? The Sigma300/350 core is very powerful and is very well suited for processing FFTs and BiQuad filters. 

Here is how the core works. 

The system clock that drives the core is derived from the PLL output. Usually the PLL output is setup to be 294.912MHz. I will not go into that setup but this is the usual output with a 12.288MHz master clock. 

The core will process one instruction for each cycle. Therefore 294,912,000 instructions per second. Divide this by the sample rate and you get the number of instructions per sample period. For 48kHz fs this results in 6,144 instructions per sampler period. 

The DSP core has a quad MAC and is able to retrieve eight 32 bit data words, perform four 64bit x 32 bit multiplies and store the results in four different accumulators all in one cycle. 

This means that it can calculate four FIR taps in one cycle. So if you use 6,000 of the 6,144 cycles in a 48k fs sample period to only calculate an FIR filter, then it can calculate 24,000 taps in one sample period. 6,000*4=24,000

This means that there is not much room to do anything else but it is technically possible. 

The other thing to keep in mind when comparing to a Cortex M4, is that all of the sample period overhead for interrupts and context switching and setting up the sampling rate structure is all handled by hardware in the SigmaDSP. In the cortex it is done in software. 

The sample rate converters in the SigmaDSP are in hardware not software and there are many hardware accelerators for commonly used functions like divide, sine/cosine/tan, etc. This makes the SigmaDSP very powerful. We recently did a comparison to an M3 HiFi core and found that the SigmaDSP had 4X the processing power once all the overhead was taken into account. 

To your first question about duck noise cancellation. 

This processor does not have the super low latency of our processors that are designed for ANC like the ADAU1787. However, often the noise in a duct is lower frequency so as long as your application needs do not extend to higher frequencies, then you could use the 1466 for duct noise cancelation. I cannot give you a frequency range since it does depend on the latency of the converters and the sampling rate. There will be around a 5 or 6 sample delay from the serial input port, through the core to the serial output ports. See if this would meet your needs.

Dave T

Hi Dave,

I wonder why this as you mentioned strong DSP has high delay but ADAU1787 whi is weak:

• Up to 50 MIPS performance

Have low latence!? It seems it must be reversed in delay, doesn't it?