# Delta Sigma Modulator Data Converter with Half-Band Filter for Decimation

This example shows how to use the Delta Sigma Modulator (DSM) data converter building block with a downstream decimation filtering scheme to be used for an Analog-to-Digital converter (ADC) application. DSMs use oversampling technique which results in the phenomenon of noise shaping by which the in-band quantization noise is strongly attenuated. In general, the noise transfer function (NTF) is given by where N is the order of the DSM and D(z) is the denominator polynomial to be designed. The procedure to systematically design a realizable NTF is explained in .

The sampling rate Fs in a DSM is typically much greater than the Nyquist rate. The ratio of Fs to Nyquist rate is called Oversampling Ratio (OSR) given by: where Fb is the highest frequency content (bandwidth) of the input signal.

DSM output is at the high rate of Fs. Decimation filters are used to reduce the data rate to the Nyquist rate (i.e. 2xFb) and filter the data stream. DSM and the decimation filter together comprise the DSM ADC. This example goes through the case of a DSM using 1-bit quantizer, but it is not uncommon to design DSMs with multi-bit quantizers. In , SQNR for a second order single bit modulator is shown empirically to be close to 75dB, resulting in an Effective Number Of Bits (ENOB) of 11 bits as explained in the next section. Theoretical SNR calculation of a DSM has been explained in .

### Design Flow

A typical design flow for modeling a DSM for a specific application is as follows:

1) Design an NTF to achieve a given system specification. The order N depends on the required SQNR and the denominator polynomial D(z) is designed to increase the modulator stability at high frequencies by reducing the gain (low-pass behavior) at high frequencies. (In this example we try to achieve SQNR of 70dB by choosing a 2nd order DSM with OSR = 64. ENOB for this case is 11.3 bits using the expression given below. 2) Choose a DSM structure that provides the best trade-off for the given specifications e.g., quantizer saturation vs. input signal range. A CIFB structure is chosen in this example.

3) Design the decimation filter to be used in conjunction with the DSM to build a DSM ADC. The decimation filter bit-width is chosen to be an integer greater than or equal to ENOB.

4) Run time-domain simulation of the model and check its performance (DSM output spectrum, SNR, etc.)

5) Add impairments to see if specifications are still met or if the architecture needs to be modified. Few key impairments for DSM are switched-capacitor kTC noise, switched-capacitor clock jitter, amplifier thermal noise, etc. To simulate the DSM ADC example, open the model "DSM_Decimation_filter" attached to this example:

open_system('DSM_Decimation_Filter.slx'); An analog input (sine wave of 990Hz) stimulus is over-sampled by the DSM block and is converted into a digital bit-stream. Specification for the DSM block is shown below:

1) Sampling Frequency (Fs) = 128KHz 2) Bandwidth (Fb) = 1KHz 3) Over Sampling Ratio (OSR) = 64

### Delta Sigma Modulator

DSM block from the Mixed-Signal Blockset is a masked subsystem containing variant subsystems to define different DSM architectures and orders. Using the block you can select out of 4 different architectures with orders ranging from 2 to 6. Hence, a total of 20 different architectures can be modeled. DSM architectures can be classified as:

1) CIFB: Cascade of Integrators, Feedback

2) CRFB: Cascade of Resonators, Feedback

3) CIFF: Cascade of Integrators, Feedforward

4) CRFF: Cascade of Resonators, Feedforward

The coefficients 'a', 'g', 'b' and 'c' for the generalized DSM structures can be determined using the procedure explained in chapter 4 in . The gain coefficients in this example were obtained using the functions provided in the Delta Sigma Toolbox by Richard Schreier .

### Decimation Filter

A decimation filter serves two purposes: 1) Filter the out-of-band noise 2) "Decimate" the DSM output data rate from Fs to the Nyquist rate 2xFb = (Fs / OSR).

For efficiency of implementation, a decimation filter is implemented as a cascade of digital filters instead of a single stage. In this example, the decimation filter is designed to have an alias attenuation of > 100 dB and < 0.1 dB of passband variation. Since the OSR of DSM is 64, a decimation factor of 64 is implemented. The filter comprises a cascade of 5 sinc filters each with a decimation factor of 2, a compensation FIR filter to compensate for droop in the sinc filters, followed by a half-band filter. The order and gain of the sinc filters are mentioned in Table 14.2 in . See chapter 14 in  for a detailed discussion of Decimation Filter design.

Half-band filter is a low pass filter which reduces the maximum bandwidth of sampled data by one octave. In this example, half-band filter has been implemented using FIR Decimation block with the filter coefficients obtained by using the DSP System Toolbox function 'firhalfband' as shown below:

firhalfband('minorder', 0.45, 1E-05); Alternatively, decimation filter can also be implemented using an FIR Decimation block (from DSP System Toolbox) with the decimation option set to 64 in the mask. More information on FIR Decimation block can be found on the documentation page for dsp.FIRDecimator.

### Simulation Results

The ADC model with 2nd order CIFB DSM block was simulated with a sinusoidal signal at 990Hz (very close to Fb). DSM output (128Ksps) is fed to a Spectrum analyzer to analyze power spectrum of the ADC output and the spectrum shows a spike at 990Hz as expected. The input to the decimation filter gets decimated by a factor of 64. The data rate is equal to 2Ksps, since OSR = 64. Spectrum of the decimation filter output (2ksps) is shown below with a spike at 990Hz. Additional simulation was run with an input frequency of 50Hz and the DSM ADC output for this sinusoidal signal is captured. The ADC output was fed to an "ADC AC measurement" block to check the performance of the DSM ADC. Both filter outputs ("FIR Decimation" and "Cascade of sinc filters") show an ENOB = 11 bits as expected.

### References

 Shanthi Pavan; Richard Schreier; Gabor C. Temes, Understanding Delta-Sigma Data Converters, second edition, IEEE Press, copyright 2017.

 Valeri Mladenov; Panagiotis Karampelas; Georgi Tsenov and Vassiliki Vita, Approximation Formula for Easy Calculation of Signal-to-Noise Ratio of Sigma-Delta Modulators

 Richard Schreier (2022). Delta Sigma Toolbox (https://www.mathworks.com/matlabcentral/fileexchange/19-delta-sigma-toolbox), MATLAB Central File Exchange. Retrieved June 14, 2022.