Strange result from small addition to an existing project

Hi Bob,

There is a small piece of framework code in the ADAU1701 (in fact, in all SigmaDSPs) that runs at the start of each processing cycle. I wonder if this code is actually preventing the ADAU1701 from being able to execute all 1024 instructions within the span of a single frame.

By the way, since version 3.0 or so, the termination cell is not required for unused outputs. Removing the terminators will probably not buy you anything, but it will help to reduce screen clutter.

Hi Brett,

Thanks for your insight on high-MIPS projects. To be safe I'll try to stay at 1000 out of 1024 instructions, although a few more may still work.

Just in case a SigmaStudio programmer at ADI is sufficiently curious, I attached two small test projects. The second one (with the "5" in its name) was directly made from the first by deleting a few blocks, right away restoring by undo, then saving with the new name. The first one (named with the "4") produces white noise when loaded into a -1701MINZ, the other one doesn't. The two projects compile to similar but non-identical DSP code, confirmed by examining the export files. Though not a serious difficulty, understanding why this happens may prove helpful to their work in some way.

Bob

Thanks very much Bob. I realize that my first response might provide some anecdotal insight but not a real solution to the problem. I will look at the projects you attached and, with the help of a SigmaDSP programmer, try to figure out what exactly is the source of the problem.

Brett,

A quick update: I did redraw the whole thing which proved worthwhile anyway. It worked fine as expected. When I added the T junction, gain, and adder shown below left, it started hissing again. When instead adding only the parametric filter on the right, it was fine:

I'm not sure yet but the latter arrangement may do what's needed, so I'm back in business. So this may become moot except for curiosity as to why this happens. I'm very familiar with both curiosity and hissing (see my avatar).

Bob

Hi Bob,

I just thought of something you might be able to try in order to see if this is actually related to your signal flow or if it's somehow related to being close to the limit of available instructions in the DSP.

Would you be able to take the hissing version from your most recent post and delete some unrelated cells from that project (for example, if there are any processing blocks in a different audio signal chain)? If you could reduce the program instruction count significantly but the hiss remains, we can conclude that it's something in the arrangement of the cells that is causing the problem, rather than the fact that you're up next to the instruction count limit.

Hi Brett,

My difficulty does not seem to be related to project size. The project without the white noise has a few more instructions (999) than the one that hisses (996). Also, removing some 100 instructions didn't help. The bad one (above, left) was an attempt to introduce a simple pole-zero (or lag-lead) compensation to the loop. This I know how to figure but it caused the hissing problem. The other one that works substitutes a parametric filter set to slightly boost the response above the critical frequency. That one is beyond my ability to analyze -- instead I used the simulation stimulus/probe to obtain the desired gain and phase response. It's working well.

Bob, thanks for the detailed analysis. This puts me at a little more ease because at least it's not pointing to some kind of SigmaStudio-side compiler issue or a SigmaDSP-side "MIPS-over" condition that might be occurring. It's probably related to your PLL design (which, I have to admit, I don't really understand all that well!).

Brett,

Thanks for having a look at this one. It appears that the answer to mixing the first-order filter with the straight gain to produce lag-lead compensation is the punch line to, "Doctor, it hurts when I do that!"

Hi Brett,

I just discovered that substituting a homemade first-order filter for the stock one stops the white noise described above. In the image below, the left circuit I had originally used causes hissing in some cases, while the equivalent filter on the right does not.

Bob

Hi Bob,

Thanks for the debug work. Just to confirm, this is running on an ADAU1701, correct?

Hi Brett,

Yes, it's a ADAU1701. Where I work we like the -1701 for the same reason we like single-chip microcontrollers -- the high-speed digital stuff with its layout and EMI issues stays inside. That's why I'd like to see a Sigma DSP with both a high-end core and a 2-in, 4-out internal codec similar to that inside the -1701.

Bob