Just a short post adding on to the OS converter thoughts. There are four basic sections of an OS A/D. The modulator and its noise shaping, the low pass filter, and the decimation filter. The first two can be folded into one section -- which can be called the "sigma - delta modulator". The task of this modulator is to generate, ( in the usual case) a one bit output stream, which is a very high frequency and dense bit stream. Since the converter is a one bit converter, in the absence of the noise shaping, the quantization noise would be horrendous, and the output would be useless.
However, the noise shaper causes the noise spectrum to become a high pass spectrum. The net result is that the in-band quantization noise is significantly lowered. As the oversampling rate increases, the in band noise comes down. As a result the SNR increases. ( e.g. a 96 dB SNR can be said to be the equivalent of 16 bits. Note that this is not an exact equivalence). The following parameters provide an indication for a fast assessment of modulator performace. The highest order of the modulator considered is 3. Third order and higher order modulators have instability issues.
For a second order modulator: The SNR is 100dB for an OSR of 192; 80dB for an OSR of 64; 60dB for an OSR of 14.
For a third order modulator, the numbers are: 100dB at OSR of 56; 80dB for an of OSR of 32; 60dB for an OSR of 14.
Note as order increases SNR increases for lower OSR.
Tuesday, February 10, 2009
Thursday, February 5, 2009
Sigma Delta data converters
As the semiconductor world moves to smaller and smaller geometries the challenge to mix analog functionality with large digital blocks looms large. A significant need is to implement data converters ( Analog to digital and digital to analog ) on large digital signal processors. I note that the engineering community is moving towards the use of oversampled data converters( sometimes known as delta - sigma or sigma - delta converters). Low order OS converters ( second order) have been available for quite sometime in standalone configurations.
As I started using these techniques myself I was surprised to note how counter intuitive the concept of the OS converter is. Understanding successive approximation and flash converters is fairly intuitive. However, OS converters are not.
Another interesting fact about OS converters is the merging of analog and digital techniques. The analog part of the converter is small in size( but perhaps not small in impact) while as far as size is concerned the digital parts are large and complex. In order to design an OS converter one needs a deeper understanding of signal processing, than for other types of data converters. First of all the OS converter contains significant digital content and is quite complicated. For the A to D we have the decimation filter ( digital) following the modulator. For the D to A we have an interpolating filter ( digital) following the modulator.
For the purely analog designer, this may cause some issues in implementation. In addition to these strange architectural constructions, there is also a whole new aspect of the noise spreading
constructs.
I recently I took a look, purely out of curiosity, at how much information on these types of devices was available in the learned journals ( I took a look at only one: IEEE Journal of Solid State Circuits) as well as the web. The following are the numbers I found:
JSSCC: From 1990 to 2009 the total number of easily accessible papers were: 18। These were the most readable and packed with information and understandable with not too much theory.
The Web yielded the following harvest।
106,000 hits for sigma - delta A to D converters, 92,000 for sigma - delta D to A converters
104,000 for digital decimation filters, 387,000 for digital interpolation filters। The surprise was 6,320,000 hits on sigma - delta modulators and finally 225,000 hits for noise shaping networks। The interest is intensse!
How does one determine what one needs as far as the parameters of a sigma delta A/D are for a particular application?
Basic parameters are sampling rate, order of the modulator, resolution in bits.
These parameters depend on the technology chosen for implementation, which determines supply voltages ( 1.8V to 350V!!), sampling frequencies ( 20khz to 1 ghz or more?), and the complexity of the analog modulator design.
My next post describes the critical parameters and their impact.
As I started using these techniques myself I was surprised to note how counter intuitive the concept of the OS converter is. Understanding successive approximation and flash converters is fairly intuitive. However, OS converters are not.
Another interesting fact about OS converters is the merging of analog and digital techniques. The analog part of the converter is small in size( but perhaps not small in impact) while as far as size is concerned the digital parts are large and complex. In order to design an OS converter one needs a deeper understanding of signal processing, than for other types of data converters. First of all the OS converter contains significant digital content and is quite complicated. For the A to D we have the decimation filter ( digital) following the modulator. For the D to A we have an interpolating filter ( digital) following the modulator.
For the purely analog designer, this may cause some issues in implementation. In addition to these strange architectural constructions, there is also a whole new aspect of the noise spreading
constructs.
I recently I took a look, purely out of curiosity, at how much information on these types of devices was available in the learned journals ( I took a look at only one: IEEE Journal of Solid State Circuits) as well as the web. The following are the numbers I found:
JSSCC: From 1990 to 2009 the total number of easily accessible papers were: 18। These were the most readable and packed with information and understandable with not too much theory.
The Web yielded the following harvest।
106,000 hits for sigma - delta A to D converters, 92,000 for sigma - delta D to A converters
104,000 for digital decimation filters, 387,000 for digital interpolation filters। The surprise was 6,320,000 hits on sigma - delta modulators and finally 225,000 hits for noise shaping networks। The interest is intensse!
How does one determine what one needs as far as the parameters of a sigma delta A/D are for a particular application?
Basic parameters are sampling rate, order of the modulator, resolution in bits.
These parameters depend on the technology chosen for implementation, which determines supply voltages ( 1.8V to 350V!!), sampling frequencies ( 20khz to 1 ghz or more?), and the complexity of the analog modulator design.
My next post describes the critical parameters and their impact.
Subscribe to:
Comments (Atom)