January 2004
Slide 1
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Practical MIMO Architecture Enabling Very High Data Rates
Jaekyun Moon, Hui Jin and Yong LiUniversity of Minnesota([email protected])
Taehyun Jeon, Heejung Yu and Sok-Kyu LeeETRI
January 13, 2004
January 2004
Slide 2
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Objectives
Identify MIMO-OFDM Architecture that allows
• 200+ Mbps data rates (10+ bps/Hz) • Practical implementation options• Backward compatibility to .11a/g
January 2004
Slide 3
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Outline• Background & motivation
– MIMO based on space-time coding– MIMO based on BICM+spatial mux
• Transmitter/receiver architecture• PER simulation results and SNR requirements• Preamble format• CFO compensation • Channel estimation
January 2004
Slide 4
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
STC vs. SM: backgroundSTC: Coded Modulation (CM)
Space-time trellis coding (STTC) ~ [Tarokh’98]
Space-time block coding (STBC) ~ [Alamouti’98], [Tarokh’99]
Space-frequency coding (SFC) (STC+OFDM) ~ [Paulraj’00], [Lu’00]
Group space-time-frequency block coding (GSTFC) ~ [Liu’02]
TCM + STBC ~ [Gong’02]
CM
SM: Bit-Interleaved Coded Modulation (BICM)BICM + Spatial multiplexing ~ [Tonello’00], [Duman’01], [Lu’02], [Park’03] S/PBICM
January 2004
Slide 5
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
STC vs. BICM+SM: philosophy STC
00/ 0
11/ 5
2
0
13
4
5
6
7
symbol
symbol
bit
BICM+SM
010
000
001011
100
101
110
111
symbolS/P
00/ 000
11/ 101
bitbitbit
January 2004
Slide 6
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
STC vs. BICM+SM: summary• STC offers improved reliability through explicit
enhancement of diversity gain– Suffers rate loss– Data rate increase possible with higher order constellation
• BICM+SM allows transmission of independent information streams, leading to high data rates– No explicit attempt to explore diversity– Yet capacity-achieving
• For high data rate applications, BICM+SM is the way to go– SNR advantage in achieving the same data rate– RF implementation difficulties associated with large constellations
(phase sensitivity, PA backoff requirements, etc)
January 2004
Slide 7
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
BICM+Spatial Multiplexing: Transmitter
3 Tx antenna example
CCencoder
symbolmapping
u c 'cπ
IFFT
IFFT
IFFT
S/P
one (parallel) OFDM symbol period
spans this Tx array containing 64x3 modulation symbols (or 64x3xm coded bits, where m is the number of coded bits per modulation symbol).
With 1 Tx antenna, the architecture coincides with that of .11a.
π
Interleaver gain in iterative demapping/decoding (IDD) is smaller here than the case where the interleaver spans the whole packet, but latency and storage requirements are reduced.
January 2004
Slide 8
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
BICM+Spatial Multiplexing: Receiver
FFT
FFT
FFT
3 Rx antenna example
one (parallel) OFDM symbol period
de-mapping
u
+
+
π1π − decoder
Note the baseband latency target of 10 us (< SIFS=16 us), a serious ASIC implementation challenge.BB latency: time that takes for the end of Rx packet that’s passing ADC to reach PHY/MAC interfaceSIFS: time between the end of Rx packet reaching Rx antenna and the start of ACK packet reaching Tx antenna
∼
January 2004
Slide 9
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Suboptimal Soft-demapper with SDF3 Tx & 3 Rx antenna example
11 21 31
12 22 32
13 23 33
H H HH H HH H H
′ ′ ′ ′ ′ ′ ′ ′= +
′ ′ ′
r a nQR decomposition
12
13 23
1 0 01 0
1HH H
= +
r a n
: modulation symbol: FFT output′
ar 1 1 1
2 2 12 1 2
3 3 23 2 13 1 3
r a nr a H a nr a H a H a n
= += + += + + +
Algorithm:Truncated-depth MAP with soft decision feedback (SDF)
January 2004
Slide 10
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Simulation Conditions• Convolutional code, punctured from (133, 171)• Random or 11a-like structured interleaver, spanning one
(parallel) OFDM symbol transmission period• Gray mapping• 64-pt FFT, 20 MHz channelization• 512 byte packet• Channel: quasi static fading, exponentially decaying power
profile with 50 ns rms delay spread, uncorrelated channel coefficients
• Soft demapping, BCJR decoding, iteration of soft decisions between demapper and decoder (IDD)
• SNR defined as the overall Tx energy (distributed to all Txantennas) to noise ratio
January 2004
Slide 11
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
)1( =τBICM+Spatial Mux: ¾ CC, 16QAM, MAP+SDF demapping
January 2004
Slide 12
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
)1( =τBICM+Spatial Mux: ¾ CC, 64QAM, MAP+SDF demapping
January 2004
Slide 13
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
)1( =τBICM+Spatial Mux: ¾ CC, 16/64QAM, MAP+SDF demapping
IDD with 3 iterations in all MIMO cases
2161444 x 4
162108*3 x 3
108722 x 2
54361 x 1
64QAM3/4 CC
16QAM3/4 CC
Ant.
January 2004
Slide 14
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Possible High Data Rates
2161444 x 4
162108*3 x 3
108722 x 2
54361 x 1
64QAM,3/4 CC
16QAM,3/4 CC
Tx-Rx antenna
802.11a data rates 6 possible .11n data rates (Mbps)
Rate information can implicitly specify the number of Tx antennas.
January 2004
Slide 15
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Other Possible High Data Rates
192964 x 4
144723 x 3
96482 x 2
48241 x 1
64QAM,2/3 CC
16QAM,1/2 CC
Tx-Rx antenna
January 2004
Slide 16
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Ref. curve: coding & interleavingover whole packet
)1( =τHow many iterations are needed?
BICM+Spatial Mux: ¾ CC, 16QAM, MAP+SDF demapping
January 2004
Slide 17
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Ref. curve: coding & interleavingover whole packet
)1( =τHow many iterations are needed?
BICM+Spatial Mux: ¾ CC, 64QAM, MAP+SDF demapping
January 2004
Slide 18
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Observations • 4x4 BICM+SM with suboptimal MAP+SDF enables
200+Mbps.– Achieves target PERs at SNRs lower than that required for 802.11a
54 Mbps data rate (1x1)– Suboptimal finite depth MAP with SDF– 3 IDD iterations used with bit-level interleaver spanning one Tx
array (Nt parallel OFDM symbols) rather then whole packet– Even 1 iteration provides substantial gain
• Once IDD is employed, the off-the-shelf CC performs comparable to turbo/LDPC codes.
• Simulation results may be somewhat optimistic in that channel coefficients are uncorrelated, but validate BICM+SM for high data rate application.
January 2004
Slide 19
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Preamble Structure, Channel Estimation and CFO Correction
January 2004
Slide 20
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Training Symbol Structure (for channel/CFO estimation)
Tx1
Tx2
Tx3
Tx4
64-12=52 frequency bins
• Both the training sequence (one or two OFDM symbols) and signal field (one OFDM symbol) are transmitted in cyclic fashion. The signal field symbol is also BPSK based and enjoys high SNR, and thus can be used for channel estimation purposes.• Fine CFO estimation can be done within one training symbol w/o CSI.• For each transmitted modulation symbol, the receiver can estimate Nr channel coefficients. After transmission of one OFDM symbol or 52 modulation symbols, the receiver estimates 52 x Nr = 208 channel coefficients, enough for channel characterization of Nt x Nr x 13 =16 x 13 = 208 coefficients or 13 temporal taps/channel.• Note the receiver does not need to know the number of Tx antennas for initial CFO/channel coefficient characterization based on training sequence and signal field.
January 2004
Slide 21
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Preamble Structure: Option 1
10 0.8 8 sµ× =
10 short symbolspreamble
1 long symbolpreamble
BPSK
Signal Field Service+Data 1 Data 2
BPSK, R=1/2
0.8 3.2 4 sµ+ = 0.8 3.2 4 sµ+ = 0.8 3.2 4 sµ+ = 0.8 3.2 4 sµ+ =
Signal detectAGCCoarse CFO estimationFrame synchronization
Fine CFO estimationChannel estimation
Rate, Length(no. of Tx antennas)Channel estimation
Channel estimation
Single Tx antenna or cyclic use of multiple
Tx antennas
Cyclic use of multiple Tx antennas
Spatial multiplexing over multiple Tx antennas
• Cuts down the preamble length by 4 us• Backward compatibility with .11a relies on protection mechanism
January 2004
Slide 22
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Preamble Structure: Option 2
10 0.8 8 sµ× = 0.8 3.2 4 sµ+ = 0.8 3.2 4 sµ+ = 0.8 3.2 4 sµ+ = 0.8 3.2 4 sµ+ =0.8 3.2 4 sµ+ =
BPSK
10 short symbolspreamble
long symbol 1preamble
Signal Field Service+Data 1 Data 2
BPSK, R=1/2BPSK
long symbol 2preamble
Fine CFO estimationChannel estimation
Signal detectAGCCoarse CFO estimationFrame synchronization
Rate, Length(no. of Tx antennas)Channel estimation
Fine CFO estimationChannel estimation Channel
estimation
Spatial multiplexing over multiple Tx antennas
Cyclic use of multiple Tx antennas
Single Tx antenna or cyclic use of multiple
Tx antennas
• Backward compatibility with .11a maintained automatically
January 2004
Slide 23
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Recursive LMMSE CFO Correction
PrincipleOperates on the time-domain received samples before FFT.Sequentially and alternately estimates data/channel info and the
phase/frequency distortion term.
featuresLow complexityVirtually no extra latencyTrack phase/frequency driftsReduce phase/frequency jitter
Data/channelextractor Invertor
MDpredictor
1−Z
kp kk|q
Channel Estimation /Symbol Detection
1|*ˆ −kky kky |1*ˆ +
January 2004
Slide 24
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Estimated versus Actual CFO
0 10 20 30 40 50 60 700
0.5
1
1.5
2
2.5
3
0 10 20 30 40 50 60 700
0.5
1
1.5
2
2.5
3
k
|q1[k]|e s tim a te of |q1[k|k]|
|q4[k]|e s tim a te of |q4[k|k]|
0 10 20 30 40 50 600.088
0.09
0.092
0.094
0.096
0.098
0.1
0.102
0.104
k
Tracking symbol/channel information during acquisition
time
time
time
CFO estimate during acquisition with actual CFO=0.1
January 2004
Slide 25
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Channel Estimation
Estimate channel’s frequency response using all available resources as needed to achieve performance/complexity/latency tradeoff:
Pilot tones within each OFDM symbolOne or two BPSK training symbols (for initial estimates) – cyclic transmission One BPSK signal field – cyclic transmissionSoft symbol info from the previous iteration (channel estimation revised to
utilize soft information)
January 2004
Slide 26
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Estimated versus Real Channel Response
0 5 10 15 200
0.5
1
1.5
k0 5 10 15 20
0
0.5
1
1.5
k
h[k]h[k] estimate
h[k]h[k] estimate
Using pilots only After one BPSK training symbol
0 5 10 15 200
0.5
1
1.5
k
h[k]h[k] estimate
Soft-decision-directed, into data packets
0 5 10 15 200
0.5
1
1.5
k
h[k]h[k] estimate
After BPSK signal field symbol
January 2004
Slide 27
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
4x4 64QAM, SDF with τ=1 MAP, with channel estimation
Ref. curve: coding & interleavingover whole packet
Ref. Line:
Coding and interleaving over whole packet
January 2004
Slide 28
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Observations 2• CFO can be estimated within one OFDM training symbol.• Channel estimation can be done with the combination of
initial training symbol(s), the use of the signal field symbol, and soft-decision-directed updates during the data portion of the packet.– The long training symbol(s) and the signal field symbol are
transmitted in cyclic fashion to provide initial channel estimates for multiple parallel channels.
• Channel response drifts can also be handled with the proposed method.
• The long symbol preamble of 11a (8 us) can be reduced to 4 us.
• The .11a preamble format can be retained, allowing automatic guarantee of backward compatibility
January 2004
Slide 29
doc.: IEEE 802.11-03/0999r0
Submission Jae Moon, U of MN
Conclusions• Practical MIMO architecture that is a natural
extension of .11a has been proposed.– Conceptually simple and straightforward generalization
of .11a to incorporate spatial dimension– Capable of 10+ bps/Hz (200+ Mbps/20 MHz) while
offering range improvement*– Can retain the same preamble format as .11a, thus
automatically guaranteeing backward compatibility– Allows straightforward ASIC implementation (no
serious latency/complexity challenges)
* With 4x4 antenna configuration and uncorrelated channel coefficients; relative to 54 Mbps .11a mode