CELLULAR COMMUNICATIONS DSP Intro
Signals: quantization and sampling
Signals are everywhere Encode speech signal (audio compression) Transfer encode signals using RF signal (modulation) Detect antenna signal Pack several calls into a single RF signal from the antenna (multiple access) Improve faded signal (equalization) Adjust transmitted signal power to save battery
What is signal? Continuous signal Real valued-function of time x=x(t), t=0 is now, t<0 is the past Can’t work with it in the computer But easy to analyze Discrete signal A sequence s=s(n), n=0 is now Values are quantized (e.g. 256 possible values) Need a time scale: n=1 is 1ms, n=2 is 2 ms etc. Can process by computer (finite portion a time)
Discrete signal from continuous Sampling Sample value of a continuous signal every fixed time interval Quantization Represent the sampled value using fixed number of levels (N=255)
Example:sampling
Example
Frequency Domain
*almost* any wave from sine waves
Frequency domain Can decompose *almost* every signal into sum of sinusoids multiplied by a *weight* Frequently domain=*weights* of sinusoids Example: Upper case letter for frequency domain X(0)=0,X(1)=1,X(2)=0.4,X(3)=0 X is the spectrum of x
Example: Sawtooth Frequency Domain X(k)=1/k
Spectrum of sawtooth
Example: Box X(n)=1/n (n is odd), X(n)=0 (n is even)
Spectrum of a linear combination Spectrum of x1+x2 is Spectrum of x1+ Spectrum of x2
Frequency Domain *Almost* every good periodic function can be represented by Two series (numbers) describe the function Recall Taylor expansion (polynomial base) Discreet Fourier Transform takes function and gives it’s Fourier representation Inverse DFT….
Representing Fourier Series Coefficient of cosines and sinus Cosine amplitude and phase Still two series, not convenient
Complex Representation Goal: single series. Trick: complex numbers Euler identity Negative frequency Complex conjugate: Two complex coefficients Complex coefficients represent real signal
DFT summary Can go back and forth from time-domain to frequency domain representation Can be computed efficiently (FFT) Signal Power in frequency and time domain (Parseval theorem)
Sampling theorem
Periodic Sampling Discrete signals are obtained from continuous signals (acoustic/speech, RF) by sampling magnitude every fixed time period How much should sampling period be for obtaining a good idea about the signal Too much samples: need more CPU, power, clock etc.
Ambiguity problem
Ambiguity Sample Frequency: Digital sequence representing also represent infinitely many other sinusoids
Aliasing Suppose our signal is composed of sinusoids from 1kHz to 4KHz (with varying weights) At sampling rate of 5 kHz we can discard 1kHz+5kHz and 4+5kHz as we know that signal has only up to 5kHz At sampling rate of 2kHz we can distinguish between 1kHz and 3kHz which both are possible
Ambiguity in frequency domain
Nyquist sampling frequency Signal band Avoid aliasing Nyquist sampling frequency Maximum frequency without aliasing
Sampling low pass signals A signal is within the known band of interest But contains some noise with higher frequencies (above Nyquist frequency) Spectrum of digital signal will be corrupted
Low Pass Filter
Time vs. Frequency Domain
Spectrum of the pulse
Time vs. Frequency Short pulse in time domain->wide spectrum
Power Spectral Density(PSD)
PSD and Separation of signals
Discrete systems
Discrete System Example:
Operation with signals Can add and subtract two signal Graphical representation
Summation
Linear Systems Simple but powerful Easy to implement
Example Example 1Hz+3Hz sine waves
Frequency domain vs. Time Domain Analyze a discrete system in time domain What it does to the sequence x(n) Analyze a discrete system in frequency domain What it does to the spectrum Change in coefficient of various sinusoids of a signal
Example:1Hz+3Hz
Nonlinear Example: 1Hz+3Hz f(x1+x2)!=f(x1)+f(x2)
Non-linear systems Might introduce additional sinusoids not present in input Results from interaction between input sinusoids Difficult to analyze Sometimes are used in practice We stick to linear systems for a while
Time-Invariant Systems Has no absolute clock Example:
Example
Unit Time Delay
Time-Delay Feasible system can’t look into a future at n=0 can’t produce x’(0)=y(4) only at n=4, can output x’(0)=y(4)
LTI: Linear Time Invariant LTI is easy to analyze and build. Will focus on them
Analyzing LTI systems
LTI systems Linear Time-Invariant Recall linear algebra A vector space has basis vectors Linear operator completely defined by its behavior on basis vectors LTI need to specify only on a single basis vector
Vector Space of Signals Shifted Unit Impulse(SUI) signal Basis for representation of the digital signals
SUI are a basis
Representation
Impulse response For time invariants systems For linear systems
Finite Impulse Response Filter Impulse response
Infinite Impulse Response
Convolution with Finite Impulse Change Index
LTI system The output of the LTI system is the result of the convolution between the input and the impulse response
Convolution
Convolution in Frequency Domain x(t), y(t) are signals X(f), Y(f) are their spectrum What is the spectrum C(f) of Convolution theorem C=X*Y (multiplication) Convolution in the time domain===Multiplication in the frequency domain
What LTI does to a signal Y=X*H Dump some sinusoids (|H(f)|<1) Boost other sinusoids (|H(f)|>1) Change phase of some sinusoids Never adds sinusoids that does not existed in the input signal
Example: Moving average
Example: 3 points weighted
Example: simple avg,more points
Magic 16 points filter