band-limiting acoustic tx
in [DSP]
|
Prev: Which type of volatile RAM has the least duration of data remanence when power-offed?
Next: Which type of volatile RAM has the least duration of data remanence when power-offed?
From: Manny on 13 May 2010 00:22 Hi, I have an acoustic transducer with fairly simple analog frontend to which I'm building a digital modulating logic. It's custom stuff so we can't rely on anything OTS (e.g. interpolation reconstruction filter) and we don't have board space to do this in analog. So basically I need to make everything happen in the digital domain. It so happens that I have plenty of on-chip ram. So I end up using a very oversampled DSO, modulate this, and raised-cosine filter quite cheaply (custom digital datapath). All is ok until here. With oversampling, the normalized frequency becomes quite large w.r.t the band of interest ruling out FIR for bandpass filtering despite being ideal group delay-wise. I end up using a low order elliptic IIR with relaxed stop bands to combat phase non-linearities there. The design as is now gives me round 25dB cleanness in the band of interest which is roughly commensurate with the SPL level of the transducer and its noise floor. What I don't like is that attenuation outside my band is not as good as I would like it to be, although in all fairness things were worked out relying somewhat on the natural response of the piezo material which we expect to do the rest. I don't want to do any fancy rate change. I'm wondering whether I can do better with what I've got at the moment. Any comments would be much appreciated. -Momo
From: Vladimir Vassilevsky on 13 May 2010 00:55
Manny wrote: > Hi, > > I have an acoustic transducer with fairly simple analog frontend to > which I'm building a digital modulating logic. It's custom stuff so we > can't rely on anything OTS (e.g. interpolation reconstruction filter) > and we don't have board space to do this in analog. So basically I > need to make everything happen in the digital domain. > > It so happens that I have plenty of on-chip ram. So I end up using a > very oversampled DSO, modulate this, and raised-cosine filter quite > cheaply (custom digital datapath). All is ok until here. > > With oversampling, the normalized frequency becomes quite large w.r.t > the band of interest ruling out FIR for bandpass filtering despite > being ideal group delay-wise. I end up using a low order elliptic IIR > with relaxed stop bands to combat phase non-linearities there. The > design as is now gives me round 25dB cleanness in the band of interest > which is roughly commensurate with the SPL level of the transducer and > its noise floor. What I don't like is that attenuation outside my band > is not as good as I would like it to be, although in all fairness > things were worked out relying somewhat on the natural response of the > piezo material which we expect to do the rest. > > I don't want to do any fancy rate change. I'm wondering whether I can > do better with what I've got at the moment. Multirate processing would be the right thing to do here. However, if you don't want to go multirate, you can optimize elliptic IIR for equiripple phase/magnitude in the passband. This would work better then any of the standard designs. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com |