I am a total newbie in signal processing. Is it possible to decimate a signal from 16.6 Khz to 16 Khz? I have an ADC interfaced to a Blackfin processor. Blackfin processor generates the CNV frequency for the ADC at 200 Khz. I need to sample data from 12 channels of the ADC, so it turns out to be 200/12 = 16.6 Khz. But in my program, the later stages require data sampled at 16 Khz. I cannot change the CNV frequency to 192 Khz in Blackfin side because of timing limitations. Is there any solution for this? Thanks a lot in advance Shafi
Decimation
Started by ●April 5, 2011
Reply by ●April 5, 20112011-04-05
On 2011/04/04 23:57, Muhammed Shafi wrote:> I am a total newbie in signal processing. Is it possible to decimate a > signal from 16.6 Khz to 16 Khz? I have an ADC interfaced to a Blackfin > processor. Blackfin processor generates the CNV frequency for the ADC > at 200 Khz. I need to sample data from 12 channels of the ADC, so it > turns out to be 200/12 = 16.6 Khz. But in my program, the later stages > require data sampled at 16 Khz. I cannot change the CNV frequency to > 192 Khz in Blackfin side because of timing limitations. > > Is there any solution for this?If you have the option of designing the hardware platform, you could build a 192 kHz clock into your circuit and run the ADC from this external clock. Brian Willoughby Sound Consulting
Reply by ●April 5, 20112011-04-05
On 04/05/2011 02:57 AM, Muhammed Shafi wrote:> I am a total newbie in signal processing. Is it possible to decimate a > signal from 16.6 Khz to 16 Khz? I have an ADC interfaced to a Blackfin > processor. Blackfin processor generates the CNV frequency for the ADC > at 200 Khz. I need to sample data from 12 channels of the ADC, so it > turns out to be 200/12 = 16.6 Khz. But in my program, the later stages > require data sampled at 16 Khz. I cannot change the CNV frequency to > 192 Khz in Blackfin side because of timing limitations. > > Is there any solution for this? > > Thanks a lot in advance > ShafiYes, it is quite possible to resample your signals. Do a search on "multirate resampling". It will require a lot of filter coefficients, but it shouldn't take a whole lot of computation to run the filter. Basically, you'll need to upsample/interpolate by 80, and then downsample/decimate by 83. -- Jim Thomas Principal Applications Engineer Bittware, Inc jthomas@bittware.com http://www.bittware.com (603) 226-0404 x536 "And" is a word that should never be used at the beginning of a sentence.
Reply by ●April 5, 20112011-04-05
Jim, Muhammad needs to keep in mind that 16.6 is only an approximation. (16.7 is a better one.) The real number is 50/3. The ratio he needs is 24/25. Jerry -- Engineering is the art of making what you want from things you can get.
Reply by ●April 5, 20112011-04-05
On 04/05/2011 06:31 AM, Jim Thomas wrote:> On 04/05/2011 02:57 AM, Muhammed Shafi wrote: >> I am a total newbie in signal processing. Is it possible to decimate a >> signal from 16.6 Khz to 16 Khz? I have an ADC interfaced to a Blackfin >> processor. Blackfin processor generates the CNV frequency for the ADC >> at 200 Khz. I need to sample data from 12 channels of the ADC, so it >> turns out to be 200/12 = 16.6 Khz. But in my program, the later stages >> require data sampled at 16 Khz. I cannot change the CNV frequency to >> 192 Khz in Blackfin side because of timing limitations. >> >> Is there any solution for this? >> >> Thanks a lot in advance >> Shafi > > Yes, it is quite possible to resample your signals. Do a search on > "multirate resampling". It will require a lot of filter coefficients, > but it shouldn't take a whole lot of computation to run the filter."Sample rate converter" will also get you some hits, as will "Polyphase Filtering". Look for the "Secret Rabbit Code" site -- it's a well-regarded bit of code for sample rate conversion with (if I recall correctly) an open-source licensing option.> Basically, you'll need to upsample/interpolate by 80, and then > downsample/decimate by 83.Mathematically, yes. You won't have to actually do this at run time. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com Do you need to implement control loops in software? "Applied Control Theory for Embedded Systems" was written for you. See details at http://www.wescottdesign.com/actfes/actfes.html
Reply by ●April 5, 20112011-04-05
On 5.4.11 4:31 , Jim Thomas wrote:> On 04/05/2011 02:57 AM, Muhammed Shafi wrote: >> I am a total newbie in signal processing. Is it possible to decimate a >> signal from 16.6 Khz to 16 Khz? I have an ADC interfaced to a Blackfin >> processor. Blackfin processor generates the CNV frequency for the ADC >> at 200 Khz. I need to sample data from 12 channels of the ADC, so it >> turns out to be 200/12 = 16.6 Khz. But in my program, the later stages >> require data sampled at 16 Khz. I cannot change the CNV frequency to >> 192 Khz in Blackfin side because of timing limitations. >> >> Is there any solution for this? >> >> Thanks a lot in advance >> Shafi > > Yes, it is quite possible to resample your signals. Do a search on > "multirate resampling". It will require a lot of filter coefficients, > but it shouldn't take a whole lot of computation to run the filter. > > Basically, you'll need to upsample/interpolate by 80, and then > downsample/decimate by 83.The OP's 16.6 kHz is actually 16 2/3 kHz, so it suffices to upsample by 24 and downsample by 25, visiting 400 ksamples/s. If he's having tight on resources, it is best to forget the sample rate conversion for 12 channels. -- Tauno Voipio tauno voipio (at) iki fi
Reply by ●April 6, 20112011-04-06
Hi All, First of all let me thank you all for your valuable solutions. If I go for Upsampling( Interpolation) i will have to fit some data based on the previous data . Am i right? Since the ADC is sampling some valuable medical diagnostic data(ECG), i guess its not a good practise to do upsampling. Is there anyway to do with just downsampling or do I need to sample at even higher frequencies for downsampling to be applied? Another option is to use 200 Khz CNV frequency and do the sampling for just 192 KHz(192k samples) and for the rest 8 Khz send some junk control word to the ADC in the TX_interrupt handler. Once it's elapsed(8k samples), again come back to the previous control words. Is this feasible? Can we keep some kind of count in the interrupt handler(ISR) that lets us know that 192k and 8k samples have been sampled? My CNV is 200 Khz and the SCLK is 8 Mhz. Regards Shafi
Reply by ●April 6, 20112011-04-06
On 06.04.2011 10:27, Muhammed Shafi wrote:> Hi All, > > First of all let me thank you all for your valuable solutions. If I > go for Upsampling( Interpolation) i will have to fit some data based > on the previous data . Am i right?No, interpolation is not the same as prediction. Read http://en.wikipedia.org/wiki/Sample_rate_conversion and follow the links and references therein. The straightforward solution is the one that Tauno Voipio suggested - interpolate by 24 and decimate by 25 (or make it in two steps interpolate firstly by 6/5 and then by 4/5, which might be easier). -- Alexander
Reply by ●April 6, 20112011-04-06
On 2011/04/05 23:27, Muhammed Shafi wrote:> Another option is to use 200 Khz CNV frequency and do the sampling for > just 192 KHz(192k samples) and for the rest 8 Khz send some junk > control word to the ADC in the TX_interrupt handler. Once it's > elapsed(8k samples), again come back to the previous control words. Is > this feasible? Can we keep some kind of count in the interrupt > handler(ISR) that lets us know that 192k and 8k samples have been > sampled?No. This will not work. If you sample at 200 kHz and then simply throw away 8,000 consecutive samples periodically, that is not the same as 192 kHz sampling. At 200 kHz sampling, there are 5.00000 microseconds between each sample point, and this gives certain frequency information. 192 kHz sampling has 5.20833... microseconds between each sample point, which gives different frequency information. The 24/25 sample rate conversion that has been repeatedly suggested will recalculate 192,000 samples based on the original 200,000, but the frequency information will be correctly preserved in the process. If your analysis were purely time domain, then you could perhaps get by with dropping 8,000 samples. But if your analysis were purely time domain then you would not have a requirement for 16 kHz sampling. Brian Willoughby Sound Consulting
Reply by ●April 6, 20112011-04-06
On 04/05/2011 11:27 PM, Muhammed Shafi wrote:> Hi All, > > First of all let me thank you all for your valuable solutions. If I > go for Upsampling( Interpolation) i will have to fit some data based > on the previous data . Am i right?Yes, kind of. Strictly speaking, interpolation means that you're taking data both before and after the point that you're interpolating, which means that you have to allow some delay into your resampling process. Taking data that is strictly before the point you're generating is extrapolation (i.e. prediction), and is much more subject to corruption by noise.> Since the ADC is sampling some valuable medical diagnostic data(ECG), > i guess its not a good practise to do upsampling.Maybe. If it's so dang valuable, why can't you just sample the ADC at the right frequency? I wouldn't expect ECG data to have much frequency content above a few hundred Hz. If this is so, then you should be able to resample it fairly cleanly. Finally -- what are you going to _do_ with it at your 16kHz? If it's some analysis, then perhaps that analysis will work fine after resampling, even if the result wouldn't look perfect to a cardiologist. You should find out what the analysis is, and whether it will tolerate resampled data. You should also find out what the characteristics of the signal you're playing with -- if it's already way oversampled, then doing a bit of interpolation on it will probably not hurt at all.> Is there anyway to do with just downsampling or do I need to sample > at even higher frequencies for downsampling to be applied?Yes. In fact, you've already been told -- sample rate conversion, which uses polyphase filtering. There's not need to actually upsample then downsample -- polyphase filtering gives you the result of that upsample/downsample combination, without forcing you to actually do all the work.> Another option is to use 200 Khz CNV frequency and do the sampling for > just 192 KHz(192k samples) and for the rest 8 Khz send some junk > control word to the ADC in the TX_interrupt handler. Once it's > elapsed(8k samples), again come back to the previous control words. Is > this feasible?If you're planning on doing what I think you are, it would mess up the signal worse than sample rate conversion.> Can we keep some kind of count in the interrupt > handler(ISR) that lets us know that 192k and 8k samples have been > sampled? > > My CNV is 200 Khz and the SCLK is 8 Mhz.-- Tim Wescott Wescott Design Services http://www.wescottdesign.com Do you need to implement control loops in software? "Applied Control Theory for Embedded Systems" was written for you. See details at http://www.wescottdesign.com/actfes/actfes.html
