You are on page 1of 14

"Confidential information -- may not be copied or disclosed without permission".

Section 6

Speech Compression

PE/TRD/GR/0109 12.02/EN

February, 2000

Speech Compression

6-1

Objectives

Speech coding
"Confidential information -- may not be copied or disclosed without permission".

Upon completion of this lesson, you will be able to:

- Draw the speech transmission chain - Determine the different rates on this chain - Explain how speech is coded - Speak about various methods to improve transmission quality

PE/TRD/GR/0109 12.02/EN

February, 2000

Speech Compression

6-2

Transmission and Reception Chains


Digitizing and source coding Channel coding Source decoding Channel decoding De-interleaving

"Confidential information -- may not be copied or disclosed without permission".

Interleaving
Deciphering Ciphering Burst deformatting

Burst formatting

section 6 section 7 section 8 section 10

Modulation

Demodulation equalization

Diversity
Transmission

PE/TRD/GR/0109 12.02/EN

February, 2000

Speech Compression

6-3

Why Digitizing and Coding the Speech?

"Confidential information -- may not be copied or disclosed without permission".

SPEECH TRANSMISSION MS
BETWEEN MOBILE AND NETWORK

BSS

SPEECH MUST BE DIGITIZED AND CODED


Lower Rate

Better Quality
PE/TRD/GR/0109 12.02/EN February, 2000 Speech Compression

64 kbit/s
6-4

Speech Transmission Chain

1 10100010110001

128 kbit/s

Digitizing
"Confidential information -- may not be copied or disclosed without permission".

Coder

13 kbit/s

S7

MS
22.8 kbit/s

Information processing (Section 7)

AIR INTERFACE
22.8 kbit/s

BSS
64 kbit/s

Decoder 16 kbit/s Signaling 13 kbit/s S 7


TCU BTS
Speech Compression 6-5

PE/TRD/GR/0109 12.02/EN

February, 2000

Hybrid Coder
ANALYSIS-BY-SYNTHESIS METHOD

Original speech 2560 bits / 20 ms

"Confidential information -- may not be copied or disclosed without permission".

Randomized Long Term excitation + Prediction 47 bits / 5 ms 9 bits / 5 ms

+
-

Perceptual filter: W(z)

Excitation generator

Synthesis H(z)=1/A(z)

Synthesized speech Linear Prediction Coding LPC 36 bits / 20 ms

Choice criterion (least square)

PE/TRD/GR/0109 12.02/EN

February, 2000

Speech Compression

6-6

Enhanced Full Rate Coder


Comparison with the Full Rate coder:
l l

"Confidential information -- may not be copied or disclosed without permission".

better quality of communications

higher robustness of the communication (less transmission errors)


lower source rate (12.2 kbit/s) enabling added protection

higher complexity (x5)

PE/TRD/GR/0109 12.02/EN

February, 2000

Speech Compression

6-7

Tandem Free Operation


Comparison with the standard transmission:
l
"Confidential information -- may not be copied or disclosed without permission".

NOT AVAILABLE YET

avoids one coding and one decoding l better communications quality between two GSM users l lower rate in the network (16 kbit/s) l allows 4 communications in the same PCM TS
l

used only for calls between two mobile stations


Speech Compression 6-8

PE/TRD/GR/0109 12.02/EN

February, 2000

Discontinuous Transmission
Principle
speech needs a lot of parameters
"Confidential information -- may not be copied or disclosed without permission".

SPEECH

40%

HIGH RATE FLOW

noise needs less parameters

NOISE

60%

LOW RATE FLOW

SAVE POWER IN THE MOBILE & REDUCE THE INTERFERENCE LEVEL


b PE/TRD/GR/0109 12.02/EN February, 2000 Speech Compression

Cell a a

6-9

Discontinuous Transmission
Voice Activity Detection
4 CRITERIA DEFINE THE ENERGETIC THRESHOLD
ENERGY

"Confidential information -- may not be copied or disclosed without permission".

SPEECH SIGNAL

STATIONARITY

TRANSMISSION
DECISION

AND / OR

PERIODICITY
PREDICTION GAIN

NOISE SIGNAL

+ DTMF detection
VOICE ACTIVITY DETECTOR
COMFORT NOISE
6-10

PE/TRD/GR/0109 12.02/EN

February, 2000

Speech Compression

Discontinuous Transmission
Comfort Noise Description 20 ms 160 samples of 16 bits

2560 bits (128 kbits/s)

"Confidential information -- may not be copied or disclosed without permission".

NOISE SIGNAL

8 kHz SAMPLING
CODER after channel coding: 456 bits in 8 bursts broadcast on TCH every 480 ms.
165 bits 95 spare bits: 0(FR)or 1(EFR)

260 bits

Silence Descriptor frame

PE/TRD/GR/0109 12.02/EN

February, 2000

Speech Compression

6-11

Discontinuous Transmission
Impact on Radio Measurements

DTX_USED
"Confidential information -- may not be copied or disclosed without permission".

DTX_NOT USED

DOWNLINK: MEASUREMENT REPORT


UPLINK: MEASUREMENT RESULT

RXLEV_SUB_SERVING_CELL RXLEV_FULL_SERVING_CELL RXQUAL_SUB_SERVING_CELL RXQUAL_FULL_SERVING_CELL


RXLEV_SUB_UP RXQUAL_SUB_UP RXLEV_FULL_UP RXQUAL_FUL_UP

The BTS measures RxLevUL_SUB RxLevUL_FULL RxQualUL_SUB RxQualUL_FULL And sets a DTX flag

The MS measures RxLevDL_SUB RxLevDL_FULL RxQualDL_SUB RxQualDL_FULL And sets a DTX flag DTX_USED or DTX_NOT USED
Speech Compression 6-12

DTX_USED or DTX_NOT USED


PE/TRD/GR/0109 12.02/EN February, 2000

Check Your Learning

1- What is the bit rate after sampling for Full Rate Speech?
"Confidential information -- may not be copied or disclosed without permission".

2- How does the coder work?

3- Characterize the tandem free operation

4- How does the DTX work?

5- What is the comfort noise and why is it used?

PE/TRD/GR/0109 12.02/EN

February, 2000

Speech Compression

6-13

"Confidential information -- may not be copied or disclosed without permission".

PE/TRD/GR/0109 12.02/EN February, 2000 Speech Compression 6-14