Hi there! I've got a problem with my 8th order IIR filter. In fact, it's a elliptic high pass designed with matlab. Cutoff frequency is 350Hz, attenuation 40dB and sample rate 8000Hz. Pass band ripple was designed to 0.1dB, but after transforming it into biquad structure and programming it, the ripple is about 1.5dB. That is too much for me. Only 0.5dB are acceptable. Has anyone some sugestions for me? Thanks, Christian.
How to achieve a lower pass band ripple?
Started by ●October 21, 2004
Reply by ●October 21, 20042004-10-21
Christian wrote:> Hi there! > > I've got a problem with my 8th order IIR filter. > In fact, it's a elliptic high pass designed with matlab. Cutoff > frequency is 350Hz, attenuation 40dB and sample rate 8000Hz. Pass band > ripple was designed to 0.1dB, but after transforming it into biquad > structure and programming it, the ripple is about 1.5dB. That is too > much for me. Only 0.5dB are acceptable. Has anyone some sugestions for > me? > > Thanks, Christian.Roundoff error? Insufficient word length? Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●October 21, 20042004-10-21
In article <980b0231.0410210017.79b0d968@posting.google.com>, Christian <christian.scholz@alpha-com.biz> wrote:> Hi there! > > I've got a problem with my 8th order IIR filter. > In fact, it's a elliptic high pass designed with matlab. Cutoff > frequency is 350Hz, attenuation 40dB and sample rate 8000Hz. Pass band > ripple was designed to 0.1dB, but after transforming it into biquad > structure and programming it, the ripple is about 1.5dB. That is too > much for me. Only 0.5dB are acceptable. Has anyone some sugestions for > me? > > Thanks, Christian.Hi You need to check your transformation. There should not be any noticeble difference. Regards, Lasse
Reply by ●October 22, 20042004-10-22
Seems to be no rounding error at all. Just got it. It was the hardware guy's part. The hardware itself produced the ripple, the filter was correct... :-) Greetings to all the 'softies' out there! He, he. Christian Jerry Avins <jya@ieee.org> wrote in message news:<2tq42cF21qhftU2@uni-berlin.de>...> Christian wrote: > > > Hi there! > > > > I've got a problem with my 8th order IIR filter. > > In fact, it's a elliptic high pass designed with matlab. Cutoff > > frequency is 350Hz, attenuation 40dB and sample rate 8000Hz. Pass band > > ripple was designed to 0.1dB, but after transforming it into biquad > > structure and programming it, the ripple is about 1.5dB. That is too > > much for me. Only 0.5dB are acceptable. Has anyone some sugestions for > > me? > > > > Thanks, Christian. > > Roundoff error? Insufficient word length? > > Jerry
Reply by ●October 22, 20042004-10-22
Most likely due to limited word length - your math is not the same as theory. The choices you make in grouping poles and zeroes into sections, and in ordering the sections, influence the quantization errors in practice. Rule of thumb: avoid big swings in gain between sections by pairing the pole nearest the unit cirle with the zero nearest to it - then repeat for remaining pairs. The zero tends to pull down the gain of the pole. I explain this in web pages from a while back, starting at: http://www.bores.com/courses/intro/iir/5_quant.htm The textbooks can seem misleading because they suggest that such grouping and ordering makes no difference (prove by the transposition theorem) but this does not take account of quantization errors which can be large in systems with feedback. Also, of course avoid quantization by choosing your original design well - also explained in links from the above pages. Chris =========================== Chris Bore BORES Signal Processing www.bores.com chris@bores.com christian.scholz@alpha-com.biz (Christian) wrote in message news:<980b0231.0410210017.79b0d968@posting.google.com>...> Hi there! > > I've got a problem with my 8th order IIR filter. > In fact, it's a elliptic high pass designed with matlab. Cutoff > frequency is 350Hz, attenuation 40dB and sample rate 8000Hz. Pass band > ripple was designed to 0.1dB, but after transforming it into biquad > structure and programming it, the ripple is about 1.5dB. That is too > much for me. Only 0.5dB are acceptable. Has anyone some sugestions for > me? > > Thanks, Christian.