ADAU1452 FIR filter size and delay time
At this moment we are starting a new project and evaluating several DSP
options, one of them is ADAU1452. We are planning using it but we have some
doubts
User manual shows that ADAU1452 can perform 24.000 taps per sample at 48Khz
sample rate but I have my doubts about it ,because of in the sigma studio
software the maximum number of coefficients is 800.
Please could you let me know how many taps could we use?
Also let me know the maximum delay time we could implement on DSP?
Audio Signal Processors
adau1452
SigmaDSP Audio Processors
Hello David,
This question has partially been answered here: Max FIR size for ADAU1452?
The ADAU1452 SigmaStudio session seems to allow for a maximum 10,000 taps with the FIR cell. If you need a larger size FIR filter you could always partially compute the larger FIR filter separately with smaller FIR blocks, then add the filtered results together (similar concept here to how we add the SHARC's SIMD processing element Rx/Sx register results after MAC operations are computed in parallel).
I do not think you can achieve 24000 taps even thought the instructions would support it. While the Simga300 architecture can compute 4 32-bit FIR taps in a single instruction cycle, we would run out of memory before utilizing all of the MIPS for quad-MAC operations.
Using existing SigmaStudio 3.11 FIR filter cells, I was able to to compile a combination of FIR filters which added up to about 20440 taps @ 48 kHz fs. (I found that sometimes for large FIR sizes I would get a memory exception error. To work around this I just shut down and restarted SigmaStudio).
-JT
Hi David,
Sorry, I forgot to answer you 2nd question.
The maximum "theoretical" delay at 48 kHz fs is around 853 milliseconds.
We have 2 DMx blocks (DM0 & DM1) of memory that are 20,480 x 32-bits in size.
(1 / 48000 seconds) x (20,480 words per block) x (2 blocks of memory) = .0.85333333 seconds.
Of course the audio framework will take up some memory, so the actual number for max delay will be smaller (about 80 DM words for a stereo passthru without any additional processing).
-JT
Hi David,
Silly me... sorry but the filter shown above will NOT work since it is not one continuous delay line for convolution sum. (For SHARC we are working with one continuous delay line when summing partial results in SIMD mode). But the example still demonstrates the memory is there for multiple large FIR filters, unfortunately we have a current maximum limit for 10000 words fore each FIR filter.
If you need a larger size, I can try to code up a custom module to allow for a larger FIR filter. Unless your application will use a smaller size than 10000 words.
-JT
Hello John,
Thanks for your quick replay.
Your tips are highly appreciate for us.
We are planning to use 2048 taps because of their calculation will take long time, it depends of low frequency behaviour which usually need long FIR.
Regarding delay I was asking about how many K Word could I use or could I dedicated for the delay. I did not mean about delay for each FIR filter.
Our current platform takes 6K 24.bit memory data exclusivity for the delay. It give us at 48Khz fs, 125ms delay time.
For that purpose could I use part of 40K Words RAM? or there is another best option?
Best regards
David
Hi David,
It seems you should be all set then for the FIR 2K block.
Regarding the memory numbers, the total delay pool for the ADAU1452 would be 853 msec. But in terms of IIR/FIR digital filters this would be cut in half to 20K available (since we have data and filter coefficients in both blocks). So realistically the filtering delay lines would max out at about 426 msec total available with up to 20K available for filter coefficients.
In terms of delay processing only, the current SigmaStudio 3.11 and prior only allowed you to specify a default z^-n Delay from DM1 space. The soon-to-be-released SigmaStudio 3.12.1 beta includes new support for specifying which memory block you want to place the delay, either DM0 or DM1. The nice thing about this is since delays tend to be much larger, you can move the storage to DM0 if you start running out of DM1 space. This would allow you to have up to 20K words of delay in either block. This is what the new cell looks like:
If you only need 6K of delay and not much filter post-processing, you could consider the ADAU1451 with it's 8K memory blocks, however if you still have a 2K FIR filter requirement, probably the ADAU1452 is your best bet.
-JT
Dear Daivd,
I want to evaluate ADAU1452 for 2s convolution reverber. I need connect several ADAU152 for mutiple FIR filter. But I find that 10000 TAPs FIR module only output the FIR reusult, no output 10000TAPS sample delay, which I can output this to next FIR module.
Whether ADI can provide the FIR module, which can output the Result, also the delayed original signal?
Thanks.
Jack
Hello,
Sorry for my late reply, I didn't know your request.
At this time we are still at the beginning of the project and these features are still not tested
Sorry
Keep you posted
David
This question has been assumed as answered either offline via email or with a multi-part answer. This question has now been closed out. If you have an inquiry related to this topic please post a new question in the applicable product forum.
Thank you,
EZ Admin
Please log in to reply.





