digital signal processors for mobile phone terminals - eeingrid/ee213a/lectures/ueda_class.pdf ·...
TRANSCRIPT
K. Ueda, '99 VLSI Circuits Short Course
Digital Signal Processorsfor Mobile Phone Terminals
Katsuhiko Ueda
Matsushita Electric Ind. Co., Ltd.
K. Ueda, '99 VLSI Circuits Short Course
K. Ueda, '99 VLSI Circuits Short Course
Abstract
- This talk will discuss the role of the DSP in the terminal and how to achieve high performance with low power consumption.
- Mobile phone system is moving rapidly into mobile multimedia era. A DSP architecture suitable for this new era will be also discussed.
- A DSP is one of the key components in a digital cellular phone terminal.
1
K. Ueda, '99 VLSI Circuits Short Course
Outline
1. The role of DSP in cellular phone terminal.
5. Summary.
4. Mobile multimedia DSP.
3. Issues for DSP in next generation mobile phone systems.
2. How to achieve high performance with low power consumption.
2
K. Ueda, '99 VLSI Circuits Short Course
Architecture of Portable Phone Terminal
Microcomputer Keypad/Display
-----
Role of DSP- Speech CODEC- Channel CODEC- Equalizer- Mod/Demodulator
Receiver
Synthesizer
Demodulator
Speaker
Microphone
Transmitter Modulator
Channel CODECEqualizer
CODECSpeech
3
K. Ueda, '99 VLSI Circuits Short Course
1 TDMA Frame (40ms)
125us (8KHz)
-> 2,560bits/40ms -> 64kbps(8bits*8KHz)
8bits
(u-Law)
1 slot (~6.7ms, 7kbps)
Receiver
Synthesizer
Demodulator
SpeakerMicrophone
Transmitter Modulator
Channel CODEC CODECSpeech
Speech: 138bits/40ms -> 3.45kbps
FEC: 86bits/40ms -> 2.15kbps
Total: 5.6kbps
User 1 User 2 User 3 User 4 User 5 User 6 User 1User 642kbps
Relationship between Speech sig. and Tx sig. (PDC)
Point A
Point B
Point A
Point B
4
K. Ueda, '99 VLSI Circuits Short Course
Speech CODEC
DigitizedInput Speech
Signal
Code Book Synthesis Filter
SpectrumAnalysis
Gain
Bit rate[kbps]
[MOPS]
10 200
10
20
5
1
15
PSI-CELP
13kbpsVSELP
RPE-LTP11.2kbpsVSELP
Parameterspeech signal
Minimizing the differencebetween input speech
signal and synthesized
5
K. Ueda, '99 VLSI Circuits Short Course
Dedicated DSP for Portable Phone
BASIC DSP
Dedicated
DSP
for Portable Phone
DPU
SAT
ALU
REG
(ACC)
MUL
DATA MEMORY
MEMORY X
MEMORY Y
POINTER UNIT
POINTER
ADR CTRL(ex. modulo, bit reverse)
CONTROL
INST
MEMORY
POINTER
ADR CTRL(ex. repeat, loop)
DECODER
I/O
SERI AL
PARALLEL
DMA CTRL
n- Increase Performance by adding accelerators
- Reduce Power consumption
Pc f*c*v^2 f: frequency, c: capacitance v: voltage
6
K. Ueda, '99 VLSI Circuits Short Course
A DSP Architecture for Portable Phone Terminal
Redundant binary number system
Dedicated MAC unit
Specialmemory
scheme torealize doublespeed MAC
Viterbi accelerator
A BUS 16B BUS 16
MACSAT
BSFT
DPU
INST ROM
ALU
M BUS 16
ACC
DATAREGS
RB-MAC
PU
PARALLEL
ACS
DATA MEMORY
SERIAL
I/O
IR
DEC
IP
cc
sp STACK
PROGRAMCONTROLData ROM AMA
AMB
da
CLK
GENPLL
EXT
CLK
DSP-CORE
DMA CONT
Double Access
Data RAM
Double speed MAC scheme
7
K. Ueda, '99 VLSI Circuits Short Course
16-bit 16-bit
A-BUSB-BUS
MEMORY Y
(MY)
16-bit
16-bit16-bit
EVENSIDE
ODDSIDE
MEMORY X
(MX)
16-bit
16-bit16-bit
EVENSIDE
ODDSIDE
MULTIPLIER
ACC
ADDER
32-bit
40-bit
PIPELINE REG
MAC UNIT
TEMP REG TEMP REG
POINTER Y
(PY)
POINTER X
(PX)
BARREL SHIFTER
1 MACHINE
CYCLE
1/2 MACHINE
CYCLE
1/2 MACHINE
CYCLE
D(2x)
D(2x+1)
D(2x+2)
D(2x+3)
D(2x+3)
D(2x+5)
D(2x)
D(2x+1)
D(2x+2)
D(2x+3)
D(2x+3)
D(2x+5)
D(2x) D(2x+1) D(2x+2) D(2x+3) D(2x+4) D(2x+5)
D(2y) D(2y+1) D(2y+2) D(2y+3) D(2y+4) D(2y+5)
D(2x)
*
D(2y)
D(2x+1)
*
D(2y+1)
Output of MX
TEMPREG
A-BUS
B-BUS
MULTIPLIERD(2x+2)
*
D(2y+2)
D(2x+3)
*
D(2y+3)
D(2x+4)
*
D(2y+4)
D(2x+5)
*
D(2y+5)
1 cycle
0.5 cycle
Double Speed MAC Scheme 8
K. Ueda, '99 VLSI Circuits Short Course
Accelerator for Viterbi Decoding
- Normal operation: The ALU is used as
a 16-bit processing unit.
Upper 8-bits Lower 8-bits
SHIFT REG
ALU
COMPARATOR REG
Add
Select
Compare
PM0(t-1)
PM1(t-1)
PM0(t)BMa(t)
BMb(t)
BMa(t) BMb(t)PM0(t-1) PM1(t-1)
PM0(t) = min[(PM0(t-1)+BMa(t)),
(PM1(t-1)+BMb(t))]
Two Adds, one Compare
and one Select -> ACS operation - ACS operation: The ALU is used as
two 8-bit adders.
9
K. Ueda, '99 VLSI Circuits Short Course
Comparison of the number of clock cycles needed to realizean 11.2kbps VSELP CODEC.
Effect of Accelerators
DSP w/o MAC &Viterbi Accelerators
0
20
40
60
80
100[%]
Misc
ALU
MAC
Block FloatingError Correction
- 11.4%
- 8%
Total: - 33.1%
- 9.0%
- 4.7%
DSP w/ MAC &Viterbi Accelerators
10
K. Ueda, '99 VLSI Circuits Short Course
y(0)=c(0)x(0)+c(1)x(-1)+c(2)x(-2)+c(3)x(-3)+
y(1)=c(0)x(1)+c(1)x(0)+c(2)x(-1)+c(3)x(-2)+
y(2)=c(0)x(2)+c(1)x(1)+c(2)x(0)+c(3)x(-1)+
FIR Filtering: tow outputs in parallel with delay register
Dual MAC Scheme
y(3)=c(0)x(3)+c(1)x(2)+c(2)x(1)+c(3)x(0)+
MAC1
Acc1
MAC0
Acc0
REGx(n-i+1) x(n-i)c(i) c(i)
# of MACoperations
# of Memoryreads
SingleMAC
DualMAC
Dual MACwith REG
N N N
2N 2N N
Low power consumption
11
K. Ueda, '99 VLSI Circuits Short Course
ACC
P-Reg 1
RBMU
P-Reg 2
RTBC
RBAU
RBA 1 RBA 2
RBA 3
16 b 16 b
24 b 24 b
40 b
40 b
A-BUS
B-BUS
MUL0.5 cycle
ACC0.5 cycle
RB->B CNV0.5 cycle
RBMU : Redundant Binary Multiply UnitRBAU : Redundant Binary Accumulation UnitRTBC : Redundant Binary Digit to Binary Digit Conversion Unit
0 20 40 60 80 100
ConvMmux1Preg2
Partial Product Gen.Encoder
FA Tree1(BW)RBA Tree1(RB)
Preg1 FA Tree2(BW)RBA Tree2(RB)
BW-MAC
RB-MAC
Power Consumption Ratio normalized to a BW-MAC
[%]
RB-MAC
MAC Unit using Redundant Binary Number 12
K. Ueda, '99 VLSI Circuits Short Course
Speaker
Microphone
Transmitter
Audio I/F
Keypad/Display
MicrocomputerTDMA Controller
A System LSI realizing Base Band Processing & Control
0.35 um CMOS2.5 Million9.26x10.0mm11x11mm CSP16b MCU MN102L(3MIPS)16b DSP MN1930(40MIPS)Demodulator, TDMA Controller,VCO, etc.
Features of the LSI
Process Tech.# of TransistorsDie SizePackage
Integrated IP
Receiver
Synthesizer
Demodulator
Modulator
Digital SignalProcessor
Misc VCO
13
K. Ueda, '99 VLSI Circuits Short Course
Goal of the next generation Mobile Phone System
Current System(PDC)
5.6 / 11.2 kbps
Next GenerationSystem
(W-CDMA)
8 kbps ~ 2 Mbps
Video Phone
High SpeedWirelessNetwork
High Bit Rate Data Transfer
-> MORE cycles for error correction
Video CODEC Capability
System Requirements
-> MORE data input/output to/from DSP
Of course, LOW POWER
14
Hello
K. Ueda, '99 VLSI Circuits Short Course
FreqTime
f1 f2 f3 f4
f2 A
BA B
Analog System
FDMA (Frequency Division Multiple Access)
Increasing Capability of Access Systems
f1
Freq
f1,c2
A
BTime c1c2 Digital System
ex. IS-95, W-CDMA
CDMA(Code Division Multiple Access)f1,c1
BA
Increase
- Channel Capacity - Data Speed
-> System Complexity (DSP Performance) Freq
Time
f1 f2 f3 f4
f1,s1
f1,s2
s
1
s
2
s
3
s
1
s
2
s
3
A
BA
BAB Digital System
ex. PDC, GSM, IS-54, PHS
TDMA (Time Division Multiple Access)
15
K. Ueda, '99 VLSI Circuits Short Course
DUP
Receiver
Transmitter
De-spreadA/D
D/A Spread
Rake
Channel
CODEC
Speech/
VideoCODEC
Control unit
Low-power & High speed A/D
Low-power & High speed correlator
High speed Viterbi/ Turbo decoder
Data
Voice
Video
Base band SignalProcessing unitRF unit
High speed &Low-power LSI
DSP
Next Generation Mobile Phone Terminal and Issues to LSIs 16
Low-power Video/
Audio CODEC LSI
K. Ueda, '99 VLSI Circuits Short Course
DSP Architecture for the Next Generation System
M BUS
DPP
Data
Adrs
AMA
AMB
regs Instruction
Memory
IOU
CKU
CORE
16
CKU: Clock control Unit
DPP: Direct Parallel Port
DPU: Data Processing Unit
ICU: Interrupt Control Unit
IOU: data I/O Unit
PCU: Program flow Control Unit
PU: Pointer Unit
- RF- ADC,DAC- Spreading/ Despreading
Wide bandwidth
to/from DSP
- Trace back
- Modulo
- Bit reverse
PU
PCU
ICU
High Performance Processing Unit
for Viterbi Decoding
DPU
A B
Data Memory
17
K. Ueda, '99 VLSI Circuits Short Course
Dual ACS Operation
T-1 T
PM0
PM1
PM0'
PM1'
2 path metrics are updated in 1 cycle
AU0AU1 ALUCOMP
CMPR
ASR1 ASR2
32
32
Data
Memory
PM1 PM0
{BM1,BM0}
{BM1+PM1}
{BM0+PM0}< >
{PM1,PM0}
{BM0+PM1}{BM1+PM0}
{BM0+PM1}
{BM1+PM0}< >
to Data Memory
register
{BM1+PM1}{BM0+PM0}
to Data Memory
Conventional
This scheme
8 Kbps
(VOICE)
32 Kbps
(DATA)
64 Kbps
(DATA)
20 40 60 80
Data Rate
[MIPS]
18
K. Ueda, '99 VLSI Circuits Short Course
Clock frequencyTechnologyNumber of devicesDie sizeSupply voltage
Performance
67.5MHz(14.8nsec)0.25um-CMOS(4-Metal)7,670KTr.9.41 x 9.22(=86.76)mm^21.8V(Internal), 3.3V(IO)4GOPS15frames/sec(CIF CODEC)
Features
DSP Core DSP Core
Double
Buffer
Double
Buffer
Shared
Memory
Local
MemoryLocal
Memory
DMA Controller
SDRAM
Dedicated
Engine
Dedicated
EngineVideo
I/F
Host
I/F
DA
AD
Video
Out
Video
In
CPU
Main-ProcessorSub-Processor
Shared
Register
DSP for Wireless Video Phone 19
K. Ueda, '99 VLSI Circuits Short Course
1. DSP is one of key components in portable phone terminal and high performance with low power consumption is essential factor.
2. In the mobile multimedia era, new DSPs with higher performance and increased functionality will be necessary.
3. DSP for portable phone must keep on achieving
MORE MIPS and LESS power consumption.
Summary 20
K. Ueda, '99 VLSI Circuits Short Course
[5] I. Verbauwhede and M. Touriguian, "Low Power DSP Engine for Wireless Communications," Journal of VLSI Signal Processing 18, pp.177-186, 1998.
[2] K. Honma and O. Kato, "Trends of research and development in Europe and America," Journal of The IEICE, Vol.78, no.2, pp.173-178, 1995.
[1] K. Ueda, T. Sugimura, et. al., "A 16-bit Digital Signal Processor with Specially Arranged Multiply-Accumulator for Low Power Consumption," IEICE Transaction, Vol. E78-C, No.12, pp.1709-1716, 1995.
[3] H. Kabuo, M. Okamoto, et. al., "An 80 MOPS-Peak High-Speed and Low-Power-Consumption 16-bit Digital Signal Processor," IEEE JSSC, Vol. 31, No. 4, pp.494-503, 1996.
[7] M. Okamoto K. Stone, et. al., "A High Performance DSP Architecture for Next Generation Mobile Phone Systems," IEEE DSP Workshop,1998.
[9] S. Kurohmaru, M. Matsuo, et. al., "A MPEG4 Programmable Codec DSP with an Embedded Pre/Post-processing Engine," IEEE CICC,1999.
[4] A. P. Chandrakasan, S. Sheng, et. al., "Low-power CMOS digital design," IEEE JSSC, Vol. 27, No. 4, pp.473-484, 1992.
[References]
[8] T. Ishikawa, H. Suzuki, et al., "W-CDMA hardware-related issues," IEEE ICCT,1998.
[6] N. Nakajima, H. Shibata, et al., "Baseband System LSI for Cellular Mobile Telephone," Matsushita Technical Journal, pp.46-52, 1999.
Mobile Phone for Internet & Data Communication
Panasonic P502i
Applications- E-mail- Web Browsing- Banking- Locating combining car navigation systemetc.
Ex. i Mode system provided by NTT DoCoMo