How to decide threshold for quantizing of signal detection?

[brmadhukar]

adc quantization levels

Hi,
How do I decide the threshold for quantizing for signal detection.
The MSB (sign bit) is easy, i.e., zero cross over but how about the magnitude bit?
TIA

 

[MonkeyBusiness]

Check out the following:

**broken link removed**

 

[brmadhukar]

Hi,
The link is a good one.

The SNR of a "single bit quantized" signal is supposed to decrease by 2dB (theoretically) but what I observe is far greater than 2dB on spectrum analyzer. Is the theory wrong or my observation.

 

[MonkeyBusiness]

If by "single bit quantized" you mean a sigma-delta type A2D (or D2A for that matter), then the SNR you get depends on the over-sampling ratio (i.e. the ratio between your sampling frequency and Nyquist frequency). I'm not aware of any 2dB mandatory penalty.

 

[flatulent]

2 dB

The 2 dB comes from hard limiting the signal. This reduces the SNR by 2 dB over keeping the system linear. This is from purely analog signal processing of radar and sonar type signals where you are looking for the presence or absence of a signal. It may be misapplied to the area of selecting the number of bits of an ADC.

 

[brmadhukar]

Hi flautulent,
Is there any general reference for quantizing for signal detection, effect of sampling for detection and tracking.

My application is in the area of GPS. If I were to quantize to 1 bit the received signal, I thought the response would be 2dB worse. I find it to be much more. A reference on this will definitely help.

brmadhukar

 

[flatulent]

no general ones

I do not have any specific references in mind. The 2 dB for hard limiting means that the received signal has to be 2 dB higher to get the same probability of detecting a radar or sonar return.

As far as ADC noise, the first bit gives you something like 7.8 dB or so and the rest give you 6 dB each. Since even bpsk requires about 10-15 dB snr you will have to use several bits in your ADC.

Maxim has an ap note on this **broken link removed**

 

[RegUser_2]

In general some should look at the dynamic range of the signal to be quantized. For instance if some assume a sensor with signal of 1 mVrms and noise of 10 nVrms over the sensor signal bandwidth, the dynamic range is 10^5. For an ADC to accuratelly quntize such signal, the ADC quantization noise (Q/SqRt(12)) RTI should be at least three times smaller than the sensor noise, and the maximuum input (Q*2^N) should be larger than the sensor signal:

3*Q/SqRt(12) = SNoise (or SNoise = Q) and Q*2^N = SSignal

hence

SSignal/SNoise = SNR = 2^N

hence the ADC resolution should be at least

N = ceil( log2(SNR) ) or N = ceil(20*log10(SNR)/20*log10(2))

assuming Nyquist sampling @ twice the sensor signal bandwidth.

In our example N should be at least 17 bit.

Hope this is helpfull.

 

[brmadhukar]

Hi,
The explanations are good, but what I am looking is for signal below noise power. The voltage leves in the above explanations should be swapped. In such case how do we handle quantization?