Hi, I am trying to implement an IIR High pass filter to filter out a DC component introduced from a ADC convter we are using. I set the passband frq 200Hz, stopband freq 50Hz, Rp 0.2, and Rs -30dB. Using Matlab, I got the following coefficients for using 2 2nd order biquid filter. B0 B1 B2 A0 A1 A2 0.8354 -1.6689 0.8354 1 -1.7319 0.7631 1 -1.9839 1 1 -1.9401 0.9609 After quantizing the coefficients, I implemented in the code using TI C5409 Fixed Point DSP. However, it doesn't seem to work. And for the coefficients that are larger than 1, I divide them by 2 and add them back togather later. Does anyone see anything wrong with the design? It seems to be pretty straightforward. Thanks.
HELP! IIR High Pass filter
Started by ●July 24, 2003
Reply by ●July 25, 20032003-07-25
"L Hao" <hao_li@yahoo.com> wrote in message news:Cc%Ta.238$hn1.67256@news.uswest.net...> Hi, I am trying to implement an IIR High pass filter to filter out a DC > component introduced from a ADC convter we are using. > > I set the passband frq 200Hz, stopband freq 50Hz, Rp 0.2, and Rs -30dB. > Using Matlab, I got the following coefficients for using 2 2nd orderbiquid> filter. > > B0 B1 B2 A0 A1 A2 > 0.8354 -1.6689 0.8354 1 -1.7319 0.7631 > 1 -1.9839 1 1 -1.9401 > 0.9609 > > > After quantizing the coefficients, I implemented in the code using TIC5409> Fixed Point DSP. However, it doesn't seem to work. And for thecoefficients> that are larger than 1, I divide them by 2 and add them back togatherlater.> > Does anyone see anything wrong with the design? It seems to be pretty > straightforward.Does this design work in matlab? If you tried using this filter on a test signal in matlab and the results don't show sufficient filtering, then it's your filter design. If the matlab simulation makes sense, then check your implementation in the fixed point DSP.> > Thanks. > >
Reply by ●July 25, 20032003-07-25
L Hao wrote:>...> > B0 B1 B2 A0 A1 A2 > 0.8354 -1.6689 0.8354 1 -1.7319 0.7631 > 1 -1.9839 1 1 -1.9401 > 0.9609 >... Please use a fixed-width font. Avoid tabs; use spaces instead. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●July 25, 20032003-07-25
Thanks for the response. The Matlab simulation looks good. THe problem seems to be the overflow of fixed point calculation. Any suggestions? Thanks. "Bhaskar Thiagarajan" <bhaskart@deja.com> wrote in message news:bfrki0$htlup$1@ID-82263.news.uni-berlin.de...> "L Hao" <hao_li@yahoo.com> wrote in message > news:Cc%Ta.238$hn1.67256@news.uswest.net... > > Hi, I am trying to implement an IIR High pass filter to filter out a DC > > component introduced from a ADC convter we are using. > > > > I set the passband frq 200Hz, stopband freq 50Hz, Rp 0.2, and Rs -30dB. > > Using Matlab, I got the following coefficients for using 2 2nd order > biquid > > filter. > > > > B0 B1 B2 A0 A1A2> > 0.8354 -1.6689 0.8354 1 -1.7319 0.7631 > > 1 -1.9839 1 1 -1.9401 > > 0.9609 > > > > > > After quantizing the coefficients, I implemented in the code using TI > C5409 > > Fixed Point DSP. However, it doesn't seem to work. And for the > coefficients > > that are larger than 1, I divide them by 2 and add them back togather > later. > > > > Does anyone see anything wrong with the design? It seems to be pretty > > straightforward. > > Does this design work in matlab? If you tried using this filter on a test > signal in matlab and the results don't show sufficient filtering, thenit's> your filter design. If the matlab simulation makes sense, then check your > implementation in the fixed point DSP. > > > > > Thanks. > > > > > >
Reply by ●July 25, 20032003-07-25
I have seen this problem in ADC's due to the differential gain difference in integrated ADC drivers such as the AD8138. A simple alternative to implementing a high pass filter is to simply calculate the mean value of the noise and then store this value as a constant and subtract it from each sample. This could be done once (may change a bit over temperature/time) or periodically with terminated input, or sinusoid. Kurt Vetter "L Hao" <hao_li@yahoo.com> wrote in message news:Cc%Ta.238$hn1.67256@news.uswest.net...> Hi, I am trying to implement an IIR High pass filter to filter out a DC > component introduced from a ADC convter we are using. > > I set the passband frq 200Hz, stopband freq 50Hz, Rp 0.2, and Rs -30dB. > Using Matlab, I got the following coefficients for using 2 2nd orderbiquid> filter. > > B0 B1 B2 A0 A1 A2 > 0.8354 -1.6689 0.8354 1 -1.7319 0.7631 > 1 -1.9839 1 1 -1.9401 > 0.9609 > > > After quantizing the coefficients, I implemented in the code using TIC5409> Fixed Point DSP. However, it doesn't seem to work. And for thecoefficients> that are larger than 1, I divide them by 2 and add them back togatherlater.> > Does anyone see anything wrong with the design? It seems to be pretty > straightforward. > > Thanks. > >
Reply by ●July 25, 20032003-07-25
Kurt Vetter wrote:> > I have seen this problem in ADC's due to the differential gain difference in > integrated ADC drivers such as the AD8138. A simple alternative to > implementing > a high pass filter is to simply calculate the mean value of the noise and > then > store this value as a constant and subtract it from each sample. This could > be done > once (may change a bit over temperature/time) or periodically with > terminated > input, or sinusoid. > > Kurt Vetter >Most noise sources are zero mean or can be made to be zero mean. The difficulty lies in the excursions. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●July 26, 20032003-07-26
I have tried a similar approch i.e. calculating the average value of incoming samples and substract the average from each sample. It does not work well. I also tried to save all the incoming data and calculated the average by hand, then substract the value from each sample at realtime. This one works better than the previous one, however it still doesn't work completely. The vendor also told me that the DC offset could change from chip to chip, which led to think of a high pass filter. The current high pass filter I am working on is a 4th order elliptic IIR in direct form II using TI fixed point C5409. I am a newbie on the fixed point DSP. It seems that I need to do coefficient scaling properly to avoid overflow. I am sure there have been a lot discussion on this one already, please don't mind for me to bring it up again. Any suggestion is appreciated! Thanks. "Kurt Vetter" <vetterk@optonline.net> wrote in message news:f1kUa.147874$ye5.25902154@news4.srv.hcvlny.cv.net...> I have seen this problem in ADC's due to the differential gain differencein> integrated ADC drivers such as the AD8138. A simple alternative to > implementing > a high pass filter is to simply calculate the mean value of the noise and > then > store this value as a constant and subtract it from each sample. Thiscould> be done > once (may change a bit over temperature/time) or periodically with > terminated > input, or sinusoid. > > Kurt Vetter > > > "L Hao" <hao_li@yahoo.com> wrote in message > news:Cc%Ta.238$hn1.67256@news.uswest.net... > > Hi, I am trying to implement an IIR High pass filter to filter out a DC > > component introduced from a ADC convter we are using. > > > > I set the passband frq 200Hz, stopband freq 50Hz, Rp 0.2, and Rs -30dB. > > Using Matlab, I got the following coefficients for using 2 2nd order > biquid > > filter. > > > > B0 B1 B2 A0 A1A2> > 0.8354 -1.6689 0.8354 1 -1.7319 0.7631 > > 1 -1.9839 1 1 -1.9401 > > 0.9609 > > > > > > After quantizing the coefficients, I implemented in the code using TI > C5409 > > Fixed Point DSP. However, it doesn't seem to work. And for the > coefficients > > that are larger than 1, I divide them by 2 and add them back togather > later. > > > > Does anyone see anything wrong with the design? It seems to be pretty > > straightforward. > > > > Thanks. > > > > > >
Reply by ●July 28, 20032003-07-28
"L Hao" <hao_li@yahoo.com> wrote in message news:IvjUa.1477$vX1.58836@news.uswest.net...> Thanks for the response. > > The Matlab simulation looks good. THe problem seems to be the overflow of > fixed point calculation. > > Any suggestions? Thanks. > > > "Bhaskar Thiagarajan" <bhaskart@deja.com> wrote in message > news:bfrki0$htlup$1@ID-82263.news.uni-berlin.de... > > "L Hao" <hao_li@yahoo.com> wrote in message > > news:Cc%Ta.238$hn1.67256@news.uswest.net... > > > Hi, I am trying to implement an IIR High pass filter to filter out aDC> > > component introduced from a ADC convter we are using. > > > > > > I set the passband frq 200Hz, stopband freq 50Hz, Rp 0.2, andRs -30dB.> > > Using Matlab, I got the following coefficients for using 2 2nd order > > biquid > > > filter. > > > > > > B0 B1 B2 A0 A1 > A2 > > > 0.8354 -1.6689 0.8354 1 -1.7319 0.7631 > > > 1 -1.9839 1 1 -1.9401 > > > 0.9609 > > > > > > > > > After quantizing the coefficients, I implemented in the code using TI > > C5409 > > > Fixed Point DSP. However, it doesn't seem to work. And for the > > coefficients > > > that are larger than 1, I divide them by 2 and add them back togather > > later. > > > > > > Does anyone see anything wrong with the design? It seems to be pretty > > > straightforward. > > > > Does this design work in matlab? If you tried using this filter on atest> > signal in matlab and the results don't show sufficient filtering, then > it's > > your filter design. If the matlab simulation makes sense, then checkyour> > implementation in the fixed point DSP. > > > > > > > > Thanks. > > > > > > > >Try implement fixed-point in Matlab using int(), floor(), ceil() type functions. You probably need not get bit-exact, but certainly get rid of the floating point. You can then analyse where the DSP is overflowing/saturating.
Reply by ●July 28, 20032003-07-28
"L Hao" <hao_li@yahoo.com> wrote:>Hi, I am trying to implement an IIR High pass filter to filter out a DC >component introduced from a ADC convter we are using. > >I set the passband frq 200Hz, stopband freq 50Hz, Rp 0.2, and Rs -30dB. >Using Matlab, I got the following coefficients for using 2 2nd order biquid >filter. > >B0 B1 B2 A0 A1 A2 >0.8354 -1.6689 0.8354 1 -1.7319 0.7631 >1 -1.9839 1 1 -1.9401 >0.9609 > > >After quantizing the coefficients, I implemented in the code using TI C5409 >Fixed Point DSP. However, it doesn't seem to work. And for the coefficients >that are larger than 1, I divide them by 2 and add them back togather later.you're comment about adding back together is unclear, and maybe wrong. 1) divide all coefficients by two 2) perform the filter multiplication 3) shift result left by 1, to get dividing factor back Robert www.gldsp.com ( modify address for return email ) www.numbersusa.com www.americanpatrol.com
Reply by ●July 28, 20032003-07-28
"Bhaskar Thiagarajan" <bhaskart@deja.com> wrote:>"L Hao" <hao_li@yahoo.com> wrote in message >news:Cc%Ta.238$hn1.67256@news.uswest.net... >Does this design work in matlab? If you tried using this filter on a test >signal in matlab and the results don't show sufficient filtering, then it's >your filter design. If the matlab simulation makes sense, then check your >implementation in the fixed point DSP.I have found that implementing the filter in C code can help a lot as well. I would highly recommend this. Most of the time, I will implement the filter in floating point, and then compare the results step by step with the DSP results. There will be differences, because of the precision differences, but they should not be too large, with the 32/40 bit accumulator ... and the data trends should be the same. Regards, Robert www.gldsp.com ( modify address for return email ) www.numbersusa.com www.americanpatrol.com
