Power Control and Prediction in Mobile Communications Systems

1
Power Control and Prediction in Mobile
Communications Systems

• Power control schemes are often applied in mobile
  communications systems
• To keep the received power of each mobile user at
  base station as equal as possible
• Three purposes
• Overcomes near-far effect (without power
  control)
• Prolongs battery lifetime in mobile users
• Maximize overall user capability

2
Two Multipath Signal Components Received at the
Base Station
3
Rayleigh Fading Channel Simulator
4
Fading Power Signals of A Typical Rayleigh Channel
5
Conventional Power Control in Mobile
Communications Systems
Bang-Bang Control
6
Conventional Power Control in Mobile
Communications Systems

• Radio channels are nonlinear and time-varying
• Deep fadings caused are harmful to power
  regulation
• Conventional feedback power control method
  suffers from slow response and large overshoots
  (accurate prediction is necessry)
• Compensation commands are delayed

7
Conventional Power Control in Mobile
Communications Systems

• Conventional bang-bang power control always
  yields large overshoot, long rise time, and large
  steady state error
• Fuzzy power control schemes utilize some priori
  knowledge of the dynamics of the fading channels
• Fuzzy power controllers provide better performance

8
Fuzzy Power Control in Mobile Communications
Systems Chang97
9
Fuzzy Power Control in Mobile Communications
Systems

• Fuzzy PI power controlle has two inputs
  and
• Fuzzy PI control rules always have the following
  form
• An example is as follows

10
Fuzzy PI Power Controller
11
Membership Functions of Fuzzy PI Controller
Variables
12
A Representative Fading Signal
Regions I and III Response of second order
systems
Region II Deep downward fading
13
Fuzzy Rules for Regions I and III
For Normal Fadings Only
14
Fuzzy Rules for Region II
For Deep Fadings Only
15
Fuzzy PI Power Control Performance
Overshoot Reduced Oscillation Elliminated
16
Comparison Between Fuzzy PI and Conventional
Fixed-Step Power Control
RMS of Tracking Error
17
Fuzzy Filtering

• Conventional filters, such as FIR and IIR, always
  introduce some delays in signal processing
• FIR and IIR filters are not efficient in
  nonlinear signal filtering
• Fuzzy filters can combine numerical and
  linguistic information Wang1993
• Numerical information from input/output data
• Linguistic information from experts
• Fuzzy filters are adaptive and predictive filters

18
Fuzzy Power Command Enhancement in Mobile
Communications Systems Gao1997

• Power commands in real cellular communications
  systems are always transferred in single-bit mode
• Multi-bit transmission mode is not practical
• Single-bit transmission causes delays in power
  control response
• Fuzzy logic is employed to generate enhanced
  power commands

19
Fuzzy Power Command Enhancement Unit
Mobile Station
20
Fuzzy Power Command Enhancement Unit

• Fuzzy power command enhancement unit is applied
  in the mobile station
• Fuzzy rules are derived based on four principles
• If mobile station receives consecutive large
  power commands, enhanced power command should
  also be large
• If mobile station receives consecutive small
  power commands, enhanced power command should
  also be small
• If mobile station receives consecutive increasing
  power commands, enhanced power command should be
  more increased
• If the mobile station receives consecutive
  decreasing power commands, enhanced power command
  should be more decreased

21
Fuzzy Power Command Enhancement Unit

• Fuzzy rules have the following form
• An example of fuzzy rule
• Advantages of fuzzy power command enhancement
  unit
• Simple and easy for implementation
• Can co-operate with any power control scheme at
  the base station

22
Membership Functions for Original Power Commands
23
Membership Functions for Enhanced Power Commands
24
Received Power Level with One-Bit Power Commands
Bang-Bang Control
25
Received Power Level with Fuzzy Power Command
Enhancement
Bang-Bang Control
26
Received Power Level with One-Bit Power Commands
Prediction Control with A Neuro Predictor
27
Received Power Level with Fuzzy Power Command
Enhancement
Prediction Control with A Neuro Predictor
28
Neural Networks-based Predictive Signal Filtering

• Predictive signal filtering is important in
  understanding system dynamics and compensating
  for instrumentation delays
• Conventional filters, such as FIR and IIR, always
  introduce some delays
• Neural networks-based filters are predictive
  filters

29
Neural Networks-based Predictive Filters
30
Training Phase of Neuro Predictive Filters
31
Comparison Between Polynomial and Neuro Predictors
32
An Example Power Prediction in Mobile
Communications Systems

• Neural networks-based predictors are applied to
  predict the received power at the base station
• single-step ahead prediction Gao1997
• multi-step ahead prediction with temporal
  difference (TD) method Gao1998

33
One-Step-Ahead Prediction of Fading Signal Using
Neural Networks
Dotted Desired Solid Actual
Gao1997,1998
34
Neural Networks-based Predictive Power Controller
35
Received Power Level Using Conventional
Bang-Bang Controller
Full Power Command Mode
36
Received Power Level Using Neural Networks-based
Predictive Controller
Full Power Command Mode
37
Received Power Level Using Conventional
Bang-Bang Controller
Single-Bit Power Command Mode
38
Received Power Level Using Neural Networks-based
Predictive Controller
Single-Bit Power Command Mode
More at Gao1997 and Gao1998
39
Acceleration Measurement in Motor Control Systems
with Neural Networks

• Acceleration feedback is necessary in the
  construction of servo controllers
• Velocity signals from low-cost encoders are often
  noise distorted
• Direct backward-difference approximation always
  generates unacceptable noise

40
An Example Noisy Velocity Curves of An Elevator
Car
41
Angular Acceleration Obtained Using
Backward-Difference Method
42
Acceleration Acquisition Using Predictive Signal
Processing Methods
Ovaska1998
43
Neural Networks-based Acceleration Acquisition
Scheme Gao1998
44
Errors of Measured and Filtered Velocity Signals
Measured Velocity
ANFIS is better than BP
BP Output
ANFIS Output
45
Velocity Measurement in Motor Control Systems
with Fuzzy Logic Gao1999

• Velocity feedback is necessary in the
  construction of servo controllers
• Velocity signals from low-cost encoders are often
  noise distorted
• Fuzzy filters produce predictive outputs
• Self-Organizing Map (SOM) can be applied to
  fine-tune fuzzy filters

46
DC Servo Motor System
47
A Fuzzy Logic-based Filter
ANFIS-based Filter
48
Noisy Velocity Signal
49
Evenly Distributed Fuzzy Membership Functions
50
SOM Applied in Fuzzy Filters

• Membership function centers can be optimized by
  SOM
• 1. Neurons in SOM are considered as fuzzy
  membership function centers
• 2. Applying competitive learning algorithm with
  the available training data (input signal)
• 3. Distribution of trained neurons is equal to
  the topology of membership functions
• Membership function widths are chosen manually

51
Optimized Membership Functions
52
Output of Conventional Fuzzy Filter
53
Output of Our New Fuzzy Filter
54
Residual Filtering Error of Conventional Fuzzy
Filter
55
Residual Filtering Error of Our New Fuzzy Filter
56
Soft Computing in Motor Fault Detection and
Diagnosis

• Motors are intensively applied in various
  industrial applications
• Fault diagnosis is very important in assuring
  safety of motor systems
• prevent eventual failures from happening
• save maintenance cost
• minimize downtime
• Soft computing methods are promising in new fault
  diagnosis techniques

57
Neural Networks-based Motor Fault Diagnosis
58
Motor Fault Diagnosis with GA Optimization of
Elman Neural Network Gao2000a

• Initial outputs of context nodes in Elman neural
  network play an important role in the network
  prediction accuracy
• Hybrid training of Elman neural network consists
  of two parts
• Gradient descent algorithm for weights
• Genetic algorithms for initial context nodes
  outputs

59
Elman Neural Network
60
Training Procedure of Elman Neural Network with
Pure BP Learning
61
GA-Evolved Optimization Process of Initial
Context Nodes Outputs
62
Motor Fault Diagnosis Using Elman Neural Network
with GA-aided Training
63
Soft Computing Methods in Control and
Instrumentation Other Examples

• A/D converter resolution enhancement using neural
  networks Gao1997
• Neural networks-based dynamic friction
  compensation in motor control systems Gao1999
• Linguistic motor fault diagnosis scheme
  Gao2000b, Gao2000c
• More details downloadable from
• http//www.hut.fi/Units/PowerElectronics/personne
  l/gao.html