Hi, all
I was simulated two second-order feedforward sigma delta modulators recently. the model shown in sdm_ff.jpg, the spectrum shown in fft_sdm_ff.jpg, the result is correct. When i add the input signal feedforward path to the quantizer shown in sdm_ffxin.jpg, the result can not display the second order noise shaping effect. But this structure can usually be seen in papers or other situations. But why the result is so unreasonable?? All comments are welcome. thanks in advance.
yeah, the STF has changed to 1, the spectrum figure include the input signal's amplitude well, but the NTF not changed, because the quantization noise's loop is the same. the two integrators' output swing were very small.
Indeed the integrators are processing quantization noise only, but the only difference between two models is whether the input signal participate in feedforward adding. After adding input signal's feedforward, the STF is changed to 1, but the feedback loop which deals with quantization noise is not changed, so the NTF is the same between two behavior models. The input signal's amplitude is not changed from two spectrum figure, but noise floor was go up.
Ok this is Silva's architecture (Silva-Temes). The STF with feedforward is 1. But, the NTF with the coefficeints is not really (1-z^-1)^2. I tried the original Silva architecture, and I am getting same SNR with the feedforward as well as without it. Did you plot the NTF for your case?
In fact the STF is always 1 if input is added to feedfoward at any coefficient.
For the integrator's swing, the coefficient must be adjusted, and this is will change the poles' position rather than (1-1/z)^2. In Silva's architecture, the integrator's gain is 1, this is a bit large, must do some scaledown.
caosl
I simulated your delta-sigma modulator with the coefficient values you have shown- I am not getting the SNR degradation you have mentioned. But, I think what is more important is the output swing of your first integrator- it is rail to rail. As per the feedforward architecture, your integrators are processing quantization noise only, but with single bit quantizer your quantization noise is also rail to rail and that is showing up at the output of the first integrator. I think using a single-bit quantizer may not be the wise thing to do.
hi, Bupesh
Whether you simulated it with multi-bit quantizer? What confusing me is that why it works fine if i take the input signal feedforward path off ? I think the additional path doesn't affect the NTF at all. Could you send me your matlab .m and .mdl files? thank you very much.
You get rail to rail output swing of first integrator? and i get approximate -0.1~+0.1, the input amplitude is 0.8, quantizer output is -1~+1.
hi,bupesh,
i simulated the silva 2'th order DSM you provided in the forum,
but i set the coefficient according to the paper
wideband low-distortion delta-sigma ADC topology, J.Silva U.Moon, J. Steensgaard"
the question is i found the integrator output contain input signal spectrum,
could you know what wrong happened?
Hi casol,
I think the main reason is you satuate the quantizer and the loop could not work properly.
BTW, the coefficents of loop filter is not reasonable. How can you get the paramters?