`Generation Communication Systems
`
`Dr. Lin-Nan Lee
`Vice President, Hughes Network Systems
`Germantown, Maryland 20854
`October 8, 2003
`
`1
`19/10/2003
`
`HUGHES PROPRIETARY II
`
`CALTECH - EXHIBIT 2019
`Apple Inc. v. California Institute of Technology
`IPR2017-00728
`
`
`
`Evolution of Error Coding
`Technology
`
`♦ 1st Generation Wireless – FM no coding, 10-11 dB C/N threshold
`♦ 2nd Generation Wireless - Convolutional codes with Viterbi decoding
`5-7 dB C/N threshold, depending on code rate, and constraint length,
`10-4 or better BER
`♦ DVB-S – Concatentated Convolutional codes with Reed-Solomon
`codes, 5-7 dB C/N threshold, depending on code rate, 10-9 or better
`BER
`♦ 3rd Generation Wireless – Turbo Codes, 0-1 dB Eb/No threshold at
`very low code rate (1/4, or 1/3), depending on block length 10-2
`Packet Error Rate
`♦ DVB-S2 – Low Density Parity Check Codes (LDPC), thresholds at 1 dB,
`10-7 MPEG Packet Error Rate
`
`HUGHES PROPRIETARY II
`
`9/10/2003
`
`21
`
`
`
`Highlights of LDPC Codes
`
`♦ LDPC codes are discovered by R. Gallager in the mid 60s
`Random Coding theory states that almost all randomly designed codes are
`•
`good, as long as they are sufficiently long
`Fewer ones in the parity check bits makes decoder simple to implement
`•
`Still too complex to implement in the 60s, or even 80s
`•
`♦ Due to the performance of turbo codes, whose performance is built
`upon
`Large random interleaver
`•
`Iterative decoding
`•
`♦ Neal and McKay “rediscovered” LDPC codes recently employing
`iterative decoding to achieve turbo-like performance
`♦ To design a good LDPC code, efficient use of modern Random Access
`Memory (RAM) architecture is the key. Design of the codes that has
`sufficient structure to allow efficient read/write, but still preserve
`sufficient “randomness” to retain coding gain are necessary
`♦ LDPC codes are selected as the DVB-S2 standard over 7 other turbo
`code based candidates because of its more efficient implementation as
`well as better performance
`
`HUGHES PROPRIETARY II
`
`9/10/2003
`
`31
`
`
`
`Eb/No vs Throughput
`Performance in AWGN
`11
`10
`
` DVB-S2 (HNS)
`(LDPC + BCH codes)
`
` DVB-S1
`(convolutional + RS codes)
`
`8-PSK, Channel Capacity
`
`QPSK, Channel Capacity
`
`123456789
`
`C/N (dB)
`
`♦ About 0.6-0.8 dB away
`from Shannon limit
`♦ About 0.3 dB better
`than the best turbo
`code candidates in the
`DVB-S2
`♦ About 0.7 dB better
`than turbo code based
`ASIC solutions we have
`tested
`♦ About 2.5-3.0 dB power
`advantage, or up to 30
`% through-put
`improvement over
`DVB-S
`♦ Further performance
`improvement not
`expected for decades
`to come
`
`0.5
`
`1.0
`
`2.0
`1.5
`bits/symbol
`
`2.5
`
`3.0
`
`HUGHES PROPRIETARY II
`
`9/10/2003
`
`41
`
`
`
`…LDPC Performance in AWGN
`
`QPSK
`
`8-PSK
`
`16-APSK
`
`32-APSK
`
`3/5
`
`2/3
`
`3/5
`
`9/10
`8/9
`
`3/4
`
`3/4
`
`4/5
`
`3/4
`
`4/5
`
`2/3
`
`2/3
`
`5/6
`
`5/6
`
`3/4
`
`4/5
`
`5/6
`
`5/6
`
`8/9
`9/10
`
`8/9
`
`8/9
`
`9/10
`
`1.E-01
`
`1.E-02
`
`1/2
`
`1.E-03
`
`1.E-04
`
`1.E-05
`
`Packet Error Rate
`
`1.E-06
`
`1.E-07
`
`0
`
`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`9 10 11 12 13 14 15 16
`8
`7
`Es/No (dB)
`
`HUGHES PROPRIETARY II
`
`9/10/2003
`
`51
`
`
`
`Why LDPC Codes are Strong
`Candidates for Next Generation
`Wireless Systems
`♦ Next Generation Wireless Systems typically require higher
`data throughput in a given bandwidth
`• High rate FEC with higher order modulation (e.g. cdma2000 EVDV,
`HSDPA) is required
`• Applications require high speed by default permit use of long
`blocks
`♦ Turbo codes tend to lose performance at high code rates due
`to excessive puncturing, turbo trellis codes become rather
`complex to implement for higher order modulation
`♦ DVB-S2 work demonstrated possibility of achieving close to
`Shannon limit performance over a very wide range of C/N
`♦ Typical performance is about 0.7-1.0 dB closer to Shannon
`limit than 3G turbo codes we developed earlier
`• 17-20 percent more capacity
`
`HUGHES PROPRIETARY II
`
`9/10/2003
`
`61
`
`
`
`w/ power control
`
`Turbo, FER
`
`Turbo, BER
`
`Convolutional, FER
`
`Convolutional, BER
`
`-18.0
`
`-17.0
`
`-16.0
`Ec/Ior (dB)
`
`-15.0
`
`-14.0
`
`-13.0
`
`Translates AWGN results to
`Raleigh Fading Channel
`
`w/o power control
`
`1.E+00
`
`1.E-01
`
`1.E-02
`
`1.E-03
`
`Error Rate
`
`1.E-04
`-19.0
`
`Convolutional code
`
`Convolutional code
`
`Turbo
`code
`
`Turbo
`code
`
`FER
`
`BER
`
`10.0
`
`12.0
`11.0
`Ebi/No (dB)
`
`13.0
`
`14.0
`
`1.E+00
`
`1.E-01
`
`Error Rate
`
`1.E-02
`
`1.E-03
`
`1.E-04
`9.0
`
`♦ Our previous work on turbo codes indicated that with power control, the gain of
`turbo codes over convolutional codes observed in the AWGN channel can be
`restored in the Rayleigh fading channel
`♦ We expect relative gain of LDPC over 3G turbo codes in the AWGN channel can
`also be preserved when power control is applied
`
`HUGHES PROPRIETARY II
`
`9/10/2003
`
`71
`
`
`
`Future Work
`
`♦ Considerable additional creativity needed to tailor specific
`codes to specific system architectures and operating
`environment
`♦ More work is needed to optimize LDPC codes for shorter blocks
`• Based on our experience with turbo codes, we should not rule out
`the possibility
`• Only Shannon capacity is the limit
`
`HUGHES PROPRIETARY II
`
`9/10/2003
`
`81
`
`