Evaluation of Signal Processing Resource Management
Algorithms in 3G
Markku Piiroinen
S-38.310 tietoverkkotekniikan diplomityöseminaari
7.9.2004
General Information
• Thesis is written at Nokia networks
• Supervisor: Professor Jorma Virtamo
• Instructor: Sami Lehesaari, M.Sc.
Contents• Background • Objectives of the thesis• Motivation• Methodology• Background Information
– Network Architecture (RNC, MGW)– Signal Processing (SP) and DSPs– Calls and SP-services
• Signal processing resource management• Tools• Simulations• Measurements• Results
Background• Evolution from 2G to 3G networks offers a
wide variety of new features to mobile UE
• Roughly:2G = mobile speech3G = mobile internet
• 3G offers service independent technology platform => 3G resource management is not trivial
Objectives of the thesis• The goal is to improve the signal processing
resource management in network elements:– Radio Network Controller (RNC) – Media Gateway (MGW)
• Enable resource management algorithm evaluation without real environment– evaluation in PC environment
• Develop simulation SW and tools to ease algorithm development for future signal processing services
Motivation (1/2)• The use of real target system takes a lot of
resources– time, – people, – HW, all has limited acces and availability and
costs money• Total system is not always available
– or does not work well enough to run mass tests– other processes may disturb the run
Motivation (2/2)
• Simulated time != realtime– test runs of few days can be run in one to few
hours• Debugging
– identical resouce management code is used– even a samll bug can cause a long test re-run
• Data collection is easier and more data can be collected– huge monitorings affect system performance– in simulated system it is possible to track all the
needed details
Methodology
• Signal Processing (SP) resource selection algorithm evaluation by simulation– Simulator development– Algorithm implementation– Traffic generation– Data analysis
Background Information
Network Architecture
CS CORE
3G RAN
2G BSS
Iu-CS
Mc
NbATM or IPbackbone PSTN
BSC
A
BTSIub
PS CORE
Iu-PS
Mc
BTS
BTS
Iur
MGW
RNC
RNC
MSCServer
Iub
MGW
SGSN
Radio Network Controller (RNC)
Related RNC functions:• Outer Loop Power Control (OLPC)
• Macro Diversity Combining (MDC)
• Ciphering
• Many protocols related to radio channels (UP, MAC, RLC, ...)
Media Gateway (MGW)
Related MGW functions:• Speech Transcoding (AMR, G.711, ...)
• Speech enhancements (EC, noise suppression, ...)
• Automatic level control (ALC)
• Supplementary services– DTMF (generation/detection)
– Tones
– Announcements
– Conference calls
Signal Processing and DSPs
• Most of speech and other user data manipulations in 3G is done by Digital Signal Processors (DSPs)
• System contains thousands of DSPs – even small inefficiencies cause a lot of wasted capacity
DSP Features
• Special purpose Processors
• Small memory
• Powerful in data processing
• Cheap – compared to general purpose processors– there are many of them - big multplier
Call Example (MGW)
MGW
IuTermination
NbTermination
MGW
NbTermination
PSTNTermination
DTMF Detection
IP/ATM Backbone
PSTNRAN
Call Example (RNC)
RNC RNC
Branch Branch
AMR MDC
Branch
DRNC MDC
Node B Node B Node B
Iub
Iur
Core network
Challenges Signal Processing Resource Allocation
• Different calls and related services have very different resource needs.– Nbr of needed services, delay, memory, MIPS,
communication, ...
• Resource fragmentation• Limited information at the call setup• Minimize the inter-unit communication• Additional limitations in unit selection
– Availability issues, load balancing …
• Dynamic problem
Challenges Signal Processing Resource Allocation
Unit 1 Unit 2 Unit 1 Unit 2
AfterBefore
?
Unit 1 Unit 2
Challenges Signal Processing Resource Allocation
Call 2
Call 3
time
Call 4
Call 5
Combined service sequence
Call 1
= Service setup
= End of call
Signal Processing Resource Manager (SPRM)
• The task of SPRM is resource selection for signal processing services– Global centralized resource manager
• Main tasks of SPRM– Makes the resource selection (unit selection)– Set service configuration into units (DSP+other)
Tools
• Tools are needed to:– Import the real system configuration into the
simulator– Traffic generation
• The simulator body– reads the configuration data and runs the
algorithms on given traffic
Simulator Usage
Data for analysis
Evaluated algorithms
SP service and unit capacity configuration
RNC/ MGW configuration Simulator
Traffic configuration
Simulations (1/2)
• Simulations were run with one HW configuration
• With different loads – near the system limits
• Two kind of traffic profiles were used– One call type – Mixture of all types (each call type had certain
probability)
Simulations (2/2)
• The load was set so that the system runs on the limit most of the time – There are no problems with light traffic– Problems arise only when the system is almost
out of resources
Measurements
• Call/service success rate
• Resource utilization
• Realized/offered traffic
• + Many other parameters to assure the result correctness– mean call duration,– mean inter-time (between calls and services in one call),
etc
Results
• It is not easy to make the unit selection in the case when all the units are almost full
• Small calls/services are ok but it is difficult to find resources for bigger services
Future Work
• Better algorithms(?)
• Information about call type – pooling – capacity pre-reservation
• Reorder of services
• … more than just resource management changes are needed
Questions?
Thank You!