Adaptive LMS filtering on ADAU1701.
Hi,
I am new to the SigmaDSP software and while I had been trying to build the LMS system for a while now, I had many questions which were unclear from the document describing this SigmaDSP.
1) Is this system set up as a frame based system and if so can I change the frame size?
2) Can I use the available building blocks that came with ADAU1701, an LMS algorithm and if so do you have documentation/application notes or code that describe how it is done.
Regards
Khal
Audio Signal Processors
SigmaDSP Audio Processors
adau1701
Hello,
An LMS filter was included in the SigmaStudio 3.5 beta build. There is a brief description and example signal flow available here: http://ez.analog.com/message/26410#26410
To answer your first question, the SigmaDSP is a stream-oriented processor (as opposed to block-oriented) and the frame sizes are fixed at the sample rate. It is not possible to do multi-rate processing. The SigmaDSP is intended to run at a fixed sample rate with the PLL locked to a fixed master clock frequency.
Hi BrettG,
You are quite right that the LMS is included in the list of tools with SigmaDSP 3.5 release but when I take the ADAU1701 hardware (as this is the development kit I purchased), I don't see LMS (nor Wind noise) as part of the available selection tabs.
If I want to use the built in LMS routine it seems I would have to go for ADAU176x family .
Regards
Khaldoon
Hi Khaldoon,
Yes, I'm sorry for the oversight in my response. The LMS filter makes use of specific DSP architecture advantages that are present in the ADAU176x series but not in the ADAU170x, which is why the algorithm is only available for certain SigmaDSPs.
That is what I had figured when playing with the software for a day .
Just to make sure that I understand correctly. The SigmaDSP ADAU1701, is sampled based processing based on the sampling frequency set in the system. In the functional Block Diagram of ADAU1701 hardware document (Figure 1), it mentioned 40ms Delay memory which is what through me towards Frame based processing. I am glad that it is not frame based.
Regards
Khal
Hi Khal,
Yes, the 40 ms figure is calculated based on the memory size. There are roughly 2048 data memory words for audio delay, which works out to 2048 / 48000 = 42 ms at a sample rate of 48 kHz. Of course, data memory is required for some other processing blocks as well, so realistically you'll have 40 ms or less of available delay to play with.
Please log in to reply.