# System Object Authoring for an Instantaneous Frequency Estimator

This example takes the perspective of a MATLAB developer willing to author an instantaneous frequency estimator based on a Discrete Energy Separation Algorithm. It also introduces creating System objects for custom DSP algorithms.

## Contents

- The Discrete Energy Separation Algorithms
- The Newly-developed Algorithm Used Within a Simple Test Bench
- System-level View
- Code and Architecture of
`dspdemo.RandomFMToneGenerator` - Code and Architecture of the DESA-2 Frequency-estimation Operator
- Reusability of Subcomponents -
`dspdemo.BandlimitedNoiseGenerator` - Reference

## The Discrete Energy Separation Algorithms

The Discrete Energy Separation Algorithms (or DESA) provide instantaneous estimates of amplitude and frequency for sinusoidal signals. The basic building block of DESA is the nonlinear Teager-Kaiser Energy Operator (TKEO)

In particular, the DESA-2 algorithm provides instantaneous estimates for amplitude (or AM function) and frequency (or FM function) through the following expressions

## The Newly-developed Algorithm Used Within a Simple Test Bench

To start, run an example of a final result - a simple system simulation using the newly developed DESA-2 frequency estimator (dspdemo.DesaTwo) within a simple testbench. The testbench includes an FM-modulated tone generator (dspdemo.RandomFMToneGenerator), two scopes for signal visualization and the ability to tune a number of system parameters during simulation. In particular you can experiment with changing three parameters that affect the frequency variation of the FM-modulated tone test input, i.e.

- The frequency offset
- The frequency standard deviation
- The bandwidth of the random frequency variations, identified as Frequency Volatility for short

The signal generator and the frequency estimator are both implemented as custom System objects. You can inspect the simulation script HelperDesa2Sysobj and identify these three main tasks

- Definition of the initial values of parameters
- Creation of all objects used in simulation
- Execution of the actual simulation loop

## System-level View

The simulated system could be represented graphically as follows

The diagram above helps understand how the different algorithmic components work together within the system simulation. It also helps visualize how the different portions of the system are packaged and used. In particular, all orange blocks represent System objects available within the DSP System Toolbox, while all cyan blocks represent custom MATLAB-authored System objects developed for this example.

At a high level the simulation uses two instances of the following custom System objects, both authored in MATLAB

- dspdemo.RandomFMToneGenerator, to generate the test signal
- dspdemo.DesaTwo, to estimate the frequency of the test signal

Within dspdemo.DesaTwo, notice for example the presence of two instances of dspdemo.TeagerKaiserOperator, one for and the other for

## Code and Architecture of `dspdemo.RandomFMToneGenerator`

RandomFMToneGenerator is composed of two other classes internally. One is dsp.NCO, a built-in System object, and the other is dspdemo.BandlimitedNoiseGenerator, another custom System object.

If you are familiar with UML, the following diagram gives a more detailed overview of the architecture of dspdemo.RandomFMToneGenerator:

Inspecting the code of dspdemo.RandomFMToneGenerator helps further clarify how its lower-level components work together.

The stepImpl method in particular implements the core functionality of the public step() method. Here separating responsibilities and hiding inner complexity within the other classes helps to simplify the code. Note how stepImpl calls in turn the step() methods of the two other System objects

## Code and Architecture of the DESA-2 Frequency-estimation Operator

The DESA-2 frequency-estimation operator (dspdemo.DesaTwo) is again implemented as a custom System object. This time though its step method accepts a buffer of samples as input and it returns local estimates of the tone amplitude and frequency. Comparing the equations for the operator on one hand and the class diagram on the other, notice how the DESA-2 operator is composed of two Teager-Kaiser energy operators of type TeagerKeiserEnergyOperator, as previously anticipated by visually inspecting the system-level diagram.

## Reusability of Subcomponents - `dspdemo.BandlimitedNoiseGenerator`

Packaging subcomponents as individual System objects also helps with reusing and testing.

For example a key component of dspdemo.RandomFMToneGenerator is dspdemo.BandlimitedNoiseGenerator. dspdemo.BandlimitedNoiseGenerator generates a zero-mean random signal with a prescribed bandwidth and RMS amplitude, and it is used here to define the frequency variations of the test signal around its offset value.

Similarly to the main dspdemo.DesaTwo object, dspdemo.BandlimitedNoiseGenerator can be used in isolation by creating an instance, setting its parameters and calling its step method in a simulation loop. An example simulation script is HelperDesa2SysobjTestBLNG (run).

For more information on authoring System objects for your custom algorithms, see Create System Objects

## Reference

P. Maragos, J.F. Kaiser, T.F. Quartieri, Energy Separation in Signal Modulations with Application to Speech Analysis, IEEE Transactions on Signal Processing, vol. 41, No. 10, October 1993