Can anyone point me to the reasons why IIR peaking filters become unstable at low audio frequencies? And, of course, how to get around this problem? I need a set of peaking filters that operate between 10Hz and 100Hz. Dirk
Audio Low Frequency Instability
Started by ●November 8, 2007
Reply by ●November 8, 20072007-11-08
Dirk Bruere at NeoPax wrote:> Can anyone point me to the reasons why IIR peaking filters become > unstable at low audio frequencies?The loss of numeric precision in the 2nd order IIR is proportional to Q(Fc/Fs)^2.> And, of course, how to get around > this problem?Noise shaping.> I need a set of peaking filters that operate between > 10Hz and 100Hz.Good. As you understand, this is not going to be for free :) Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by ●November 8, 20072007-11-08
On 8 Nov, 15:01, Vladimir Vassilevsky <antispam_bo...@hotmail.com> wrote:> Dirk Bruere at NeoPax wrote: > > > Can anyone point me to the reasons why IIR peaking filters become > > unstable at low audio frequencies? > > The loss of numeric precision in the 2nd order IIR is proportional to > Q(Fc/Fs)^2. >Would going to 64 bit floating point extend the frequency range? Dirk
Reply by ●November 8, 20072007-11-08
cyber.fidelity@googlemail.com wrote:> On 8 Nov, 15:01, Vladimir Vassilevsky <antispam_bo...@hotmail.com> > wrote: > >>Dirk Bruere at NeoPax wrote: >> >> >>>Can anyone point me to the reasons why IIR peaking filters become >>>unstable at low audio frequencies? >> >>The loss of numeric precision in the 2nd order IIR is proportional to >>Q(Fc/Fs)^2. >> > > > Would going to 64 bit floating point extend the frequency range?No. $1000 would do. VLV
Reply by ●November 8, 20072007-11-08
On 8 Nov, 16:12, Vladimir Vassilevsky <antispam_bo...@hotmail.com> wrote:> cyber.fidel...@googlemail.com wrote: > > On 8 Nov, 15:01, Vladimir Vassilevsky <antispam_bo...@hotmail.com> > > wrote: > > >>Dirk Bruere at NeoPax wrote: > > >>>Can anyone point me to the reasons why IIR peaking filters become > >>>unstable at low audio frequencies? > > >>The loss of numeric precision in the 2nd order IIR is proportional to > >>Q(Fc/Fs)^2. > > > Would going to 64 bit floating point extend the frequency range? > > No. $1000 would do. > > VLVNo problem. But not to you:-) Dirk
Reply by ●November 8, 20072007-11-08
cyber.fidelity@googlemail.com wrote:> On 8 Nov, 16:12, Vladimir Vassilevsky <antispam_bo...@hotmail.com> > wrote: > >>cyber.fidel...@googlemail.com wrote: >> >>>On 8 Nov, 15:01, Vladimir Vassilevsky <antispam_bo...@hotmail.com> >>>wrote: >> >>>>Dirk Bruere at NeoPax wrote: >> >>>>>Can anyone point me to the reasons why IIR peaking filters become >>>>>unstable at low audio frequencies? >> >>>>The loss of numeric precision in the 2nd order IIR is proportional to >>>>Q(Fc/Fs)^2. >> >>>Would going to 64 bit floating point extend the frequency range? >> >>No. $1000 would do. >> >>VLV > > > No problem. > But not to you:-)You see, Dirk, any number of bits can't compensate for the lack of brains. VLV
Reply by ●November 8, 20072007-11-08
cyber.fidelity@googlemail.com wrote:> On 8 Nov, 15:01, Vladimir Vassilevsky <antispam_bo...@hotmail.com> > wrote: >> Dirk Bruere at NeoPax wrote: >> >>> Can anyone point me to the reasons why IIR peaking filters become >>> unstable at low audio frequencies? >> The loss of numeric precision in the 2nd order IIR is proportional to >> Q(Fc/Fs)^2. >> > > Would going to 64 bit floating point extend the frequency range?Going from where? Incidentally, your statement of 10 to 100 Hz has no reference wothout the sampling frequency being specified. It's easy to filter in that range with a .5Khz sample rate. You can gain a few bits of significance -- those taken by the exponent -- by using well scaled fixed point. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●November 8, 20072007-11-08
On 8 Nov, 16:17, Jerry Avins <""jya\"@ieee,org"> wrote:> cyber.fidel...@googlemail.com wrote: > > On 8 Nov, 15:01, Vladimir Vassilevsky <antispam_bo...@hotmail.com> > > wrote: > >> Dirk Bruere at NeoPax wrote: > > >>> Can anyone point me to the reasons why IIR peaking filters become > >>> unstable at low audio frequencies? > >> The loss of numeric precision in the 2nd order IIR is proportional to > >> Q(Fc/Fs)^2. > > > Would going to 64 bit floating point extend the frequency range? > > Going from where? Incidentally, your statement of 10 to 100 Hz has no > reference wothout the sampling frequency being specified. It's easy to > filter in that range with a .5Khz sample rate. > > You can gain a few bits of significance -- those taken by the exponent > -- by using well scaled fixed point. > > Jerry > -- > Engineering is the art of making what you want from things you can get. > =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF==AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF= =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF Sampling frequency is a minimum of 44.1kHz Dirk
Reply by ●November 8, 20072007-11-08
Dirk Bruere at NeoPax wrote:> On 8 Nov, 16:17, Jerry Avins <""jya\"@ieee,org"> wrote: >> cyber.fidel...@googlemail.com wrote: >>> On 8 Nov, 15:01, Vladimir Vassilevsky <antispam_bo...@hotmail.com> >>> wrote: >>>> Dirk Bruere at NeoPax wrote: >>>>> Can anyone point me to the reasons why IIR peaking filters become >>>>> unstable at low audio frequencies? >>>> The loss of numeric precision in the 2nd order IIR is proportional to >>>> Q(Fc/Fs)^2. >>> Would going to 64 bit floating point extend the frequency range? >> Going from where? Incidentally, your statement of 10 to 100 Hz has no >> reference wothout the sampling frequency being specified. It's easy to >> filter in that range with a .5Khz sample rate. >> >> You can gain a few bits of significance -- those taken by the exponent >> -- by using well scaled fixed point. >> >> Jerry >> -- >> Engineering is the art of making what you want from things you can get. >> ����������������������������������������������������������������������� > > Sampling frequency is a minimum of 44.1kHzThat's not the point. A filter's critical frequencies are designed as a fraction (in the range 0 to .5) of the sample rate. Minimum doesn't cut it. You need the actual number to design the filter. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●November 8, 20072007-11-08
On Nov 8, 6:31 am, Dirk Bruere at NeoPax <dirk.bru...@gmail.com> wrote:> Can anyone point me to the reasons why IIR peaking filters become > unstable at low audio frequencies?Quantization of arithmetic leads to quantization of how much any coefficient can affect the results. Quantization of the IIR coefficients leads to quantization of the possible pole/zero locations. A pole which ends up on or outside the unit circle after its location is quantized leads to an unstable transfer function. At low peaking frequencies, usually the absolute bandwidth in relation to the sample rate is also very low. Thus the the desired poles end up close to locations which might get moved after quantization to on (nearly on) or outside the unit circle.> And, of course, how to get around > this problem?Reduce quantization errors (several methods) until all poles are guaranteed to be in stable locations. IMHO. YMMV. -- rhn A.T nicholson d.0.t C-o-M
