mcaspPlayBk with Filtering OMAPL138, C6748LCDK

I am a little bit confused what exactly do you ask,  so please

Specify what release you use

What are the functions that you call

what exactly you try to do

Provided all the above I will be happy to help

Ran

Hi Ran,

Thanks for the response.

This is with reference to the McaspPlyBk staterware code for the OMAPL138 / C6748 LCDK. I'm using the code as a base to plug in a FIR filter routine. The filter uses the tI API DSPF_sp_fir_gen. Right now i do get predicted responses after plugging in the filter routine in the DMA based staterware code. However, i do need clarifications on the questions below (at the bottom of the post). 

I'm using CCS 6.1.2 and OMAPL138_StarterWare_1_10_04_01. 

 The FIR code snippet plugged into the main() function of the McaspPlayBk.c file  is as below:

memcpy((short *)buftemp,(void *)rxBufPtr[lastFullRxBuf],
                           AUDIO_BUF_SIZE);

for(idx1=0;idx1<NUM_SAMPLES_PER_AUDIO_BUF;idx1++)
{
            inputDemoSignal[idx1+NUM_OF_TAPS-1]=(float)(buftemp[idx1*2]);
}

//Filter blockprocessing
DSPF_sp_fir_gen(inputDemoSignal,coeffsFilter,outputDemoSignal,NUM_OF_TAPS, NUM_SAMPLES_PER_AUDIO_BUF);


//Copy buffer for trasmit
for(idx1=0;idx1<NUM_SAMPLES_PER_AUDIO_BUF;idx1++){
               buftemp[idx1*2]= _spint(outputDemoSignal[idx1] * 65536) >> 16; //inputDemoSignal[idx1+NUM_OF_TAPS];   //=<<6
               dataOutBufferFull[(idx%6144)++]=buftemp[idx1*2];
        }

//Re-initialize filter state for next buffer processing
for(idx1=0;idx1<NUM_OF_TAPS-1;idx1++)//idx1<NUM_OF_TAPS
            inputDemoSignal[idx1] = inputDemoSignal[NUM_SAMPLES_PER_AUDIO_BUF + idx1];


//Copy data to transmit DMA buffer 
memcpy((void *)txBufPtr[lastSentTxBuf],(void *)buftemp, AUDIO_BUF_SIZE);

Further details regarding code and code snippet are below:

- input, filter taps and output buffer are 8 byte aligned based on the documentation

- input, filter taps and output buffer are floating point

-buffer "buftemp" is of type short and is used to receive and transmit to the DMA buffers 

- NUM_SAMPLES_PER_AUDIO_BUF has been set to 2048 

-Sampling frequency: 8000

The filter routine works fine in non-real time. When plugged into the McaspPlayBk starterware DMA based code, the input data are obtained from buffer pointed by rxBufPtr[lastFullRxBuf].  I would like clarification on the following:

1. I see that the format of data in the buffer pointed to by rxBufPtr[lastFullRxBuf]   is of the form [xxxx,0, xxxx,0, xxxx, 0, xxxx, 0...] considering 16 bit short array and "xxxx" being valid values. It's these values that i use as input to the filter routine DSPF_sp_fir_gen. Is this correct or  is there additional data interleaving of left and right channels that needs to be taken into consideration.  The rest of the code the the mcaspPlayBk  module remains "as-is" (i.e, WORDSIZE 16, SLOT_SIZE 16, NUM_I2S_CHANNELS 2)

2.  The taps that are input to the filter  DSPF_sp_fir_gen API are right now in the range [-1 to 1] floating point values. Are there any restrictions on the range of filter co-efficients to the  DSPF_sp_fir_gen routine. 

 Do let me know if you need more information.  Thanks much.

Regards

Smita

Hi Ran, 

Thanks for your inputs. 

1. Yes, i did try accessing data as "int", but all values show up positive in the range 0 to 65535 . So it seems like it's expecting 16 bit short data (Left and Right channel configurations based on configurations to the the McASP/EDMA/Codec).  Also, is the  data in rxBufPtr iin compressed form? Is the Mcasp compressing data from the codec. We would like to access raw PCM data from the codec. Any insights/inputs here would help. 

2. Yes of course, scaling filter taps would increase filter gain and correspondingly reflect in the output. I was just wondering if the DSPF_sp_fir_gen API was expecting the filter taps in a specific range. Good to know there is no restriction on range. 

Looking forward to your inputs on 1.

Regards

Smita