Signal Processing and Communications with MATLAB and Simulink
Giorgia Zucchelli Application Engineer - MathWorks
Key Takeaways
Quickly analyze and develop new algorithms with MATLAB
Accurate system-level multi-domain analysis with Simulink
With MATLAB and Simulink you can quickly design entire systems with better performance
Motivations
Quickly analyze and develop new algorithms with MATLAB
>> Evaluating innovative ideas is time-consuming
Accurate system-level multi-domain analysis with Simulink
>> Modeling implementation constraints requires specific knowledge
With MATLAB and Simulink you can quickly design entire systems with better performance
>> Optimizing the tradeoff between reuse and innovation is challenging
MATLAB for Signal Processing
Digital Filter Design Fixed-point in MATLAB and more
MATLAB for Signal Processing: Whats New
Multirate Digital Filter Design Fixed-point in MATLAB for streaming applications and more
Challenges: Digital Filter Design
During design:
Is it meeting the specs?
During implementation:
How can I minimize cost? Which filter structure will be optimal? Will it work correctly on the target hardware? What can I trade off?
Digital Filter Design in MATLAB
Design FIR and IIR filters
Many frequency responses Optimized design methods
Implement filters using various filter architectures Visualize filter response Export MATLAB filters into Simulink for system-level simulation
Digital Filter Design Tradeoffs
Example: high speed, low pass decimation filter
Specifications:
Input sampling freq = 88.4kHz Output sampling freq = 11.05kHz Decimation factor = 8 Passband ripple = 0.1dB Stopband attenuation = 90dB Passband = 3000Hz Stopband = 3200Hz
8
Takeaways: Digital Filter Design in MATLAB
Different filter responses Many optimized design methods Control of the filter architecture Evaluation of tradeoffs between performances, costs and specs Automation of the design process Rapid design iterations Visualization of filter characteristics
Challenges: Fixed-Point Design
Dynamic analysis of data range Finite word length = quantization error
Overflow (overload distortion) Underflow (granular noise)
Algorithms supporting fixed-point data type
7+1=8 bit: 7 word length & 1 fractional bit Range =[-64 63.5) Step =1/2
s 32 16 8 4 2 1 1/2 1/4 1/8 1/16 1/32
7+1=8: 7 bit word length & 5 fractional bits Range =[-4 3.9688) Step = 1/32
10
MATLAB for Fixed-Point
Represent fixed-point data type Analyze quantization effects
Built-in logging and visualizations
Accelerate execution of fixed-point code System objects for more fixed-point functions
11
New in R2010a: System Objects
MATLAB objects that represent time-based and data-driven algorithms, sources, and sinks System objects enable streaming in MATLAB Support of fixed-point data type and automatic C code generation Made available by:
Signal Processing Blockset Video and Image Processing Blockset Communication Blockset
12
MATLAB is Best at Batch Processing
All the data Work on all the data at once
Deliver all at once
13
Many Systems Demand Stream Processing
All the data Incremental delivery
All the data is not available at once
Limited memory footprint Real-time requirements
Typical applications
Communications simulation Audio / video processing Data acquisition
14
Example: Filtering of an Audio Stream
filename = 'dspafxf_8000.wav'; [audio Fs] = wavread(filename); filt = fir1(40, 0.8, 'high'); audiofilt = filter(filt,1,audio); wavplay(audiofilt,Fs);
15
Filtering in MATLAB
filename = music.wav'; [audio Fs] = wavread(filename); filt = fir1(40, 0.8, 'high'); audiofilt = filter(filt,1,audio); wavplay(audiofilt,Fs);
Must wait for all data to be processed before listening to results Loads entire dataset into workspace
audio data uses more space than needed (double vs. uint16) Overall, code uses several copies of the audio dataset in memory
16
Stream Processing in MATLAB Today
%% Streaming the MATLAB way % set up initializations filename = music.wav'; Fs = 8000; info = mmfileinfo(filename); num_samples = info.Duration*Fs; frame_size = 40; bLP = fir1(12, 0.8, low'); zLP = zeros(1,numel(bLP)-1); output = zeros(1,num_samples);
Explicit state management requires programmer to manage details that should be implicit
Indexing is tedious and %% Processing in the loop error prone index= 1; while index < (num_samples-frame_size+1) data = wavread(filename,[index index+frame_size-1]); [datafilt, zLP] = filter(bLP,1,data,zLP); output(index:index+frame_size-1) = datafilt; index = index + frame_size; end Need to maintain output buffer wavplay(output,Fs); because wavplay requires all data before being called
17
Stream Filtering with System Objects
%% Streaming with System Objects before use % set up initializations filename = music.wav'; hFilter = dsp.DigitalFilter; hFilter.TransferFunction = 'FIR (all zeros)'; hFilter.Numerator = fir1(12, 0.8, low'); hAudioSource = dsp.AudioFileReader(filename, ... 'SamplesPerFrame',40, 'OutputDataType','double'); hAudioOut = dsp.AudioPlayer('SampleRate', 11050); %% Processing in the loop while ~isDone(hAudioSource) data = step(hAudioSource); datafilt = step(hFilter, data); step(hAudioOut, datafilt); end
Initialize objects
Many ways to set object properties Source and FIR filter states are implicit Audio player runs in-the-loop with the current frame, avoiding lengthy buffer
18
In-the-loop code is much simpler
No management of indexing
Batch Processing
Load the entire video file and process it all at once
MATLAB
Ease of Use Memory
Batch Video Algorithm
Source
19 19
Traditional Stream Processing in MATLAB
Load a video frame and process it before moving on to the next frame Manually maintain indexing, buffering, states
MATLAB Ease of Use Memory
Stream Source
Stream Processing
20
Stream Processing with System Objects
Load a video frame and process it before moving on to the next frame Implicit indexing, buffering, handling of states
MATLAB Ease of Use Memory
Stream Source
Stream Processing
21
Fixed-Point Algorithms in MATLAB
Represent fixed-point data types in MATLAB as fi objects Run simulation in floating-point or fixed-point modes using data type override Log min, max and overflow
22
Takeaways: Fixed-Point System Objects
Stream processing in MATLAB
Easier to write and be correct the first time Improves handling of large data sets
Fixed-point modeling
All relevant objects support fixed-point data types Compatible with Fixed-Point Toolbox
C-code generation
Most objects support code generation using EMLC Compatible with Embedded MATLAB
23
Simulink for Signal Processing
Systems with complex timing System-level simulation and more
24
Simulink for Signal Processing: Whats New
Mixed-signal systems with complex timing System-level simulation including RF and more
25
Challenges: Mixed-Signal Systems
Anticipate physical constraints
Analog and digital electronics
Complex timing:
Continuous and discrete timing Feedback loops Threshold crossing Asynchronous behavior Concurrent paths
26
Mixed-Signal Modeling with Simulink
Design embedded systems:
Use the most suitable modeling approach Anticipate physical impairments (mixed-signal) Define the system architecture
Verify embedded systems:
Analyze close-loop behavior
27
Takeaways: Mixed-Signal Simulation
Simulation of continuous and discrete signals Multi-rate digital signals with arbitrary sample rates Complex timing
Built-in notion of concurrency Detect zero-crossings and discontinuities Enable feedback loops Asynchronously triggered blocks
Share the MATLAB workspace
28
Challenges: RF System-Level Simulation
Model RF front-ends:
Without being an expert With acceptable simulation speed
Integrate baseband and RF simulation
Develop a system-level view
29
New in R2010b: SimRF
Simulation Speed Fidelity
Circuit envelope analysis
Multi-carrier systems and arbitrary architectures
Equivalent baseband models
Single carrier super-heterodyne cascaded systems (former RF Blockset technology)
Complex baseband models
Mathematical analytical models
30
Former RF Blockset Technology
Single carrier simulation of cascaded RF systems
Linear elements are modeled with baseband-complex equivalent descriptions Nonlinear elements are described by means of (static) AM/AM AM/PM characteristics Elements are cascaded for fast single carrier simulations in the time domain
31
Modeling Paradigm: RF Blockset
From passband-real signal to baseband-complex equivalent signal
Pass-band transfer function
MHz GHz 0
fc Bandwidth = 1/Ts
Baseband-complex equivalent transfer function
-0.5/Ts
frequency
+0.5/Ts
RF Blockset Baseband equivalent time-domain impulse response
Number of sub bands (freq. resolution) equals length of impulse response
0
time
32
New SimRF Technology
Multi-carrier simulation for arbitrary topology of RF systems
Use circuit-envelope analysis for multi-carrier linear and non-linear elements
Extend in band analysis to multiple bands
Based on Simscape to model networks of physical components
Possibility to build arbitrary topologies Possibility to probe within the network
Enables extended interferers and spurs analysis at system-level
33
Simulation Paradigm: SimRF
Envelope analysis of the modulations centered on multiple carriers
Complex envelope of modulated input signals
MHz GHz 0
fc1
fc2 frequency
Circuit-envelope analysis
MHz GHz 0
fc2-fc1
fc1
fc2
fc2+fc1 frequency
SimRF Complex envelope response around selected carrier
MHz GHz 0
fc2+fc1 frequency
34
RF Modeling with Simulink
Design embedded systems:
Use the best modeling approach Anticipate physical impairments (RF) Define the system hierarchy
Verify embedded systems:
Analyze results in streaming conditions Evaluate system-level performances
35
Takeaways: RF Modeling with Simulink
Complex hierarchical model description Streaming capabilities with frame-based processing
Make use of MATLAB for matrix processing
Use Embedded MATLAB
Increase expressivity Use of legacy-code
36
Design and Implement Signal Processing Systems with MATLAB and Simulink
Algorithm design Fast simulation Architecture exploration Targeting implementation Verification and testing Rapid prototyping
37
Conclusions
Quickly analyze and develop new algorithms with MATLAB
Accurate system-level multi-domain analysis with Simulink
With MATLAB and Simulink you can quickly design entire systems with better performance
38
