Department of Electronic Systems
Random Access Protocols
for Massive MIMO
Emil Björnson
Erik G. Larsson
Linköping University
Sweden
Elisabeth de Carvalho
Jesper H. Sørensen
Petar Popovski
Aalborg University
Denmark
2016 Tyrrhenian International Workshop on Digital Communications (TIW16)
Sept 12-14, 2016
Department of Electronic Systems
CSI acquisition and data transmission in
Crowd Scenarios
Megacities
Hotspots
Machine type
communications
http://edition.cnn.com/2013/05/02/travel/london-city-airport-internet-
of-things/
3
Presentation Content
Massive MIMO: massive number of spatial degrees of
freedom
Crowds: exploit very large multiplexing gain
Uplink Training based on orthogonal pilot sequences
Length/number of orthogonal pilots limited
Pilot shortage
One solution: random access to the pilots and possibly the
data
4
CSI Acquisition in Massive MIMO
Time-division duplexing and channel reciprocity
CSI is acquired using uplink training
Exploited for downlink transmission
Orthogonal pilots
5
Orthogonal Pilots are a limited resource
Pilot Sequence length limited by:
# of pilots limited by channel coherence time
# of pilots limited by transmit power
Number of orthogonal pilots = pilot sequence length
Pilot ShortageNumber of pilot sequences is much smaller than the
number of pilot sequences
6
Traffic Burstiness
Video streaming
Social network
Cloud
Machine-type communications: crowd of devices transmitting
sporadically
time
unpredictable and intermittent traffic
Non-streaming internet applications
7
Random Pilot Access
Crowd of devices with unpredictable and intermittent traffic
Makes pilot pre-allocation very inefficient
Need for:
Scalable and efficient pilot access and
data transmission protocols.
Random Access to
Pilot sequences
Proposed solution:
8
Random Access to Pilots
Total number of terminals 𝑲 and 𝝉𝒑 orthogonal pilots available
Random pilot selection COLLISIONS
Collisions = Pilot contamination
Users select a pilot sequence uniformly at random with
probability 𝒑𝒂
Intra-Cell Pilot contamination
9
Pilot Collision
• Users with same pilot sequence: their channel cannot be
distinguished:
• Beamforming at BS based on contaminated channel estimation:
results in inter-user interference
𝑔 =
all colliders
𝑔𝑖 + 𝑛𝑜𝑖𝑠𝑒
UL Pilot
transmission
𝑝1𝑝1
10
Two kinds of approaches
Random Access to Pilots Random Access to Pilots and
Data
𝑝1𝑝1
Pilot Contention resolution
Terminal sends payload when no
pilot contention
𝑝1𝑝1
Uplink data is embarked with the
pilots
Data affected by collision-induced
interference
Collision in the pilot domain only Collision in the pilot and data
domain
11
Pilot Contamination
Suppression
Methods based on spatial separation, diversity in path loss,
timing offsets
If successful, BS can decode ID of separable users
which are admitted for DL/UL data transmission
UL Pilot transmission
12
Pilot Contamination
Detection of collision at the BS
BS sends a message to colliding devices:
try again until no collision
Avoidance
try again
New random
accessUL Pilot transmission
13
Pilot Contamination
Detection of collision at the device
BS sends a precoding pilot
1) Device detects collision
2) Device decides whether to retransmit the pilot
sequence
Avoidance
Precoded pilot
Try again
Decision:
Retransmit UL Pilot?
UL Pilot transmission No collision
(with high probability)
14
Pilot Contamination
training
𝑔 = 𝑔1 + 𝑔2 + 𝑛𝑜𝑖𝑠𝑒 w =𝑔1+𝑔2+𝑛𝑜𝑖𝑠𝑒
𝑔1+𝑔2+𝑛𝑜𝑖𝑠𝑒
y1 =𝑔1
𝐻+𝑔2𝐻+𝑛𝑜𝑖𝑠𝑒
𝑔1+𝑔2+𝑛𝑜𝑖𝑠𝑒𝑔1 + 𝑛
𝐸 𝑔𝑖2 = 𝛽𝑖
y12 ≈
𝛽12
𝛽1+𝛽2+𝜎𝑛2
y12 =
𝛽12
𝛽1Expected
Training based Estimate of 𝛽1 + 𝛽2 from 𝑦12: 𝛽𝑠𝑢𝑚
Compare 𝛽1 to 𝛽𝑠𝑢𝑚
Compare 𝛽1 to 𝛽𝑠𝑢𝑚/2
Channel
hardening
Contaminated channel estimation Precoding
STRONGER user
Retransmit the pilot sequence
Collision detection
15
Collision resolution
16
Uplink Joint Pilot and Data Transmission
𝐶𝑊𝑇2(1) 𝒑𝟐 𝐶𝑊𝑇2
(3) 𝐶𝑊𝑇2(𝐿)𝒑𝟐
𝐶𝑊𝑇3(3)
𝐶𝑊𝑇4(1)
𝒑𝟏
𝐶𝑊𝑇4(3)𝒑𝟏
𝐶𝑊𝑇3(2)
𝐶𝑊𝑇1(𝐿)𝒑𝟏
𝐶𝑊𝑇4(3)
𝒑𝟏
𝒑𝟐 𝐶𝑊𝑇4(𝐿)𝒑𝟏
𝒑2
⋯
⋯
⋯
⋯
Time slot
𝑇1
𝑇2
𝑇3
𝑇4 𝒑𝟏
𝝉𝒑
𝝉𝒖
One codeword sees an asymptotic number of:
– channel fades (small and large scale)
– pilot contamination events
For delay-tolerant communications
17
Uplink Sum Rate Bound – MRC at BS
Large scale
fading
Rate user 0 (𝐾𝑎, 𝑐0, large scale fading)
Number of
active
users
Number of
contaminators
to user 0
EEKa, c
𝐾𝑎𝜏𝑢 − 𝜏𝑝
𝜏𝑢
= Sum Rate (𝜏𝑝, 𝑝𝑎)
𝜷𝒌
𝒈𝒌~𝑪𝑵(𝟎, 𝜷𝒌𝑰)
18
Uplink Sum Rate Bound- MRC at BS
Probability of having
Ka active terminals
out of K
Probability of having c
contaminator to a given user
conditioned on Ka active users
Optimization wrt 𝜏𝑝 and 𝑝𝑎
𝐾𝑎~Binomial(𝐾, 𝑝𝑎) 𝑐|𝐾𝑎~Binomial(𝐾𝑎 − 1, 1/τ𝑝)
Channel energy
19
Heuristic Solution
Sum Rate ~ 𝑀 𝜏𝑢
Depends on the channel
energy variations
𝑝𝑎𝑜 𝐾 = 𝑎 𝜏𝑢 𝑀𝜏𝑝
𝑜 =𝜏𝑢
3
𝐶𝑊𝑇2(1) 𝒑𝟐 𝐶𝑊𝑇2
(3) 𝐶𝑊𝑇2(𝐿)𝒑𝟐
𝐶𝑊𝑇3(3)
𝐶𝑊𝑇4(1)
𝒑𝟏
𝐶𝑊𝑇4(3)𝒑𝟏
𝐶𝑊𝑇3(2)
𝐶𝑊𝑇1(𝐿)𝒑𝟏
𝐶𝑊𝑇4(3)
𝒑𝟏
𝒑𝟐 𝐶𝑊𝑇4(𝐿)𝒑𝟏
𝒑2
⋯
⋯
⋯
⋯
Time slot
𝑇1
𝑇2
𝑇3
𝑇4 𝒑𝟏
𝝉𝒑
𝝉𝒖
20
Rate~ 𝑀 𝜏𝑢
Average Sum Rate
M=400
M=100
0.5 b/s/Hz per user
1 b/s/Hz per user
Average sum rate as a function of 𝜏𝑢, K=800
21
Average Number of Active Users
M=400
M=100
𝑝𝑎𝐾 as a function of 𝜏𝑢 , 𝐾 = 800
22
Conclusions
• New services and scenarios in 5G: new way to access
the pilots and transmit the data
• Massive MIMO is a fundamental enabler for crowd MBB
and mMTC
• Creation of an efficient standard for wireless networks
based on massive MIMO technology will require a
complete re-design of the multiple-access layer.