Sign in
Search c54x
Subscribe to c54x
c54x by Keywords
5409 | 5416 | AD5 | ADC | BIOS | Boot | Booting | Bootloader | C540 | C5402 | C5409 | C5416 | CCS | Codec | DMA | Dmad | DSK | DSKPlus | Dsplib | EVM | FFT | FIR | Flash | GPIO | HPI | Initialization | Interrupt | JTAG | LOG_printf | MCBSP | RFFT | RTDX | Sampling | STLM | UART | VC540
Discussion Groups
Discussion Groups | TMS320C54x | Processing time issues on TMS320VC5402 DSK
Technical discussions about the TI C54x DSPs (including the c5401, c5402, c5402a, c5404, c5407, c5409, c5409a, c5410, c5410a, c5416, c5420, c5421, c5441, c549, c5470 and c5471).
Processing time issues on TMS320VC5402 DSK - JACOB CHACKO - Sep 5 8:20:00 2002
|
Hi, I am currently busy implementing a voice privacy system that uses the Digital Signal Processing method TFSP. i.e. Time and Frequency segment permutation. It basically divides the input digital sample signal into the different frequency components using Quadrature mirror filters. It uses a scrambler on the transmitter section and a de-scrambler on the receiver section. I don’t intend to bore you with all the maths but infact I need your assistance with regard to completing the project. The problem my fellow friends and me are experiencing is of the processing time taking longer than the sampling period of 8000 Hz. This is causing the output to become distorted. The project is being implemented on the TMS320VC5402 DSK kit and we are writing code in C using the code composer studio. I intend to perform all my processing within each sampling instant but it seems as if as soon as my processing load increases , the signal becomes distorted, which in turn implies that I am missing samples. I wonder if you could help me with regard to this problem that we are experiencing, or infact provide us with any information that might help us with the project. I appreciate your assistance in the matter. |
|
|
(You need to be a member of c54x -- send a blank email to c54x-subscribe@yahoogroups.com )
Re: Processing time issues on TMS320VC5402 DSK - wavetrain3000 - Sep 5 15:08:00 2002
|
I assume you've done rough estimates to convince yourself that the signal processing you want to do actually fits in the 8000 hz sampling intervals. Given that, I have a couple suggestions: - don't assume that "the output is distorted" because you're missing samples: put a real detector in your code (e.g. halt if the next sample arrives before the current one is processed, or if the processing takes more than about 120us) so you know directly what's going on - the DSP debugger and run-time environment can disrupt timing; if you're working via HPI the DSP gets halted at a periodic rate for polling (don't know the details, but I observed the phenomenon directly on one of my real-time systems), and any sort of run-time interaction between debugger and DSP (printf, DSP/bios communication) can be much more disruptive than you might think - if you have routines that seem much slower than you expect, and they've compiled reasonably well, recall that external memory is much slower than internal Ken Sinclair Wavetrain - Wireless computing and embedded system design |
(You need to be a member of c54x -- send a blank email to c54x-subscribe@yahoogroups.com )
Re: Processing time issues on TMS320VC5402 DSK - Benjamin Chan - Sep 5 19:53:00 2002
|
If there is peak MIPS issue in your processing, then then try use buffer processing instead of sample by sample processing, in which you can avoid peak processing time in a particular sample time. Also check what is the clocking Mhz of the C5402. If you are not using 100Mhz, you can also try set it to see if that helps. But beware of the memory wait states setup if you are using external memory. If possible, try put critical code and data buffers in onchip memory. Good luck. Ben --- JACOB CHACKO <> wrote: > Hi, > > I am currently busy implementing a voice privacy > system that uses the > Digital Signal Processing method TFSP. i.e. Time and > Frequency segment > permutation. It basically divides the input digital > sample signal into > the different frequency components using Quadrature > mirror filters. It > uses a scrambler on the transmitter section and a > de-scrambler on the > receiver section. > > I don intend to bore you with all the maths but > infact I need your > assistance with regard to completing the project. > The problem my > fellow friends and me are experiencing is of the > processing time > taking longer than the sampling period of 8000 Hz. > This is causing the > output to become distorted. > > The project is being implemented on the TMS320VC5402 > DSK kit and we > are writing code in C using the code composer > studio. I intend to > perform all my processing within each sampling > instant but it seems as > if as soon as my processing load increases , the > signal becomes > distorted, which in turn implies that I am missing > samples. > > I wonder if you could help me with regard to this > problem that we are > experiencing, or infact provide us with any > information that might > help us with the project. > > I appreciate your assistance in the matter. > _____________________________________ __________________________________________________ |
(You need to be a member of c54x -- send a blank email to c54x-subscribe@yahoogroups.com )
Re: Processing time issues on TMS320VC5402 DSK - Georgios Chatzigeorgiou - Sep 5 22:20:00 2002
|
Hi Jacob I've encountered the same problem (not with the same algorithm). There are some quidelines presented below (the order is irrelevant), which may solve your problem. 1. Use the Optimizer If you are writing C code you'd propably want to use the optimizer. There are various levels of optimization effort, try to experiment to use which best fits your app (I would suggest level 3). 2.Allocate Time Critical Code In the on-chip memory (DARAM) Furthermore, if you can arrange your code so that 0080 to 1FFF contains data, and 2000-3FFF contains executable code, the DARAM can be fully utilized. 3. Clock speed and Wait States Make sure the DSP is clocked at the highest speed (100MHz), and that the wait states for the SRAM are set to 1 (and the multiplier is off). 4. Use Assembly The DSP works better with assembly. I managed to make the polyphase filterbank of the MP3 five times faster using assembly instead of C. Parrellel load, store, multiply, add, substract can save you from a lot of cycles (not to mention the numerous application specific instructions that can save a lot of time). Hope I've helped George Chatzigeorgiou |
(You need to be a member of c54x -- send a blank email to c54x-subscribe@yahoogroups.com )