Massive MIMO in TDD wireless networks depends crucially on channel reciprocity, which can be established by calibration. Existing calibration approaches, however, have been proven to be impractical for deployment in 5G NR and 802.11. This presentation introduces terminal-assisted calibration, which is shown to overcome the drawbacks of existing approaches and to enable various massive MIMO modes in 5G NR and 802.11.
2. 2Blue Mountain Wireless, Inc.
Relevant Key Features in 5G NR
TDD (time-division duplex) takes a much more active role
Provides flexibility in downlink/uplink allocation
Various MIMO modes benefit from channel reciprocity in TDD
Beamforming
Spatial multiplexing (multi-user MIMO)
Distributed MIMO (CoMP)
TDD is to dominate in higher frequency band (> 3 GHz)
Massive MIMO
Up to 256 antennas below 28 GHz, many more in mmWave band
2D antenna array ⇒ 3D beamforming
Self-contained TDD subframes – providing a platform for
massive MIMO
data (DL)
guard
period
ACK (UL)
adaptive DL/UL allocation within 1-ms subframe
training signal for DL MIMO by channel reciprocity
3. 3Blue Mountain Wireless, Inc.
TDD/MIMO Gains and Assumptions
4× cell throughput gain
The gain is all by TDD and massive MIMO
Source: Qualcomm
TDD/MIMO gain assumes channel reciprocity
Channel reciprocity – downlink channel and uplink channel are
reciprocal
In reality, downlink and uplink channels are not reciprocal
Reciprocity holds only between broadcaster antennas and terminal
antennas
Taken into account transceiver chains of broadcasters and
terminals, downlink and uplink channels are no longer reciprocal
In practice, calibration is needed to restore channel
reciprocity and to realize the multi-fold gain
4. 4Blue Mountain Wireless, Inc.
Channel Reciprocity
With transceiver chains taken into account:
and : diagonal matrices capturing transceiver properties of
broadcasters and terminals
Reciprocity does not hold unless and are (proportional to)
identity matrices
T,0( )b f
T, 1( )mb f
R,0( )b f
R, 1( )mb f
T,0( )t f
R,0( )t f
R, 1( )nt f
T, 1( )nt f
( )fH
5. 5Blue Mountain Wireless, Inc.
Calibration
Calibration – obtaining and so downlink channel can be
derived from uplink channel
Situations where alone is sufficient
Multi-user MIMO (spatial multiplexing)
Distributed MIMO (CoMP)
Beamforming with single-antenna terminals
In beamforming with multiple-antenna terminals, both and are
needed
6. 6Blue Mountain Wireless, Inc.
Existing Art – Self Calibration
Self-calibration is the prevalent calibration method
Self-calibration means calibration among broadcasters only
Broadcaster examples
eNBs in 4G/LTE
gNBs in 5G
APs in 802.11
In self-calibration:
No terminals are involved
Can only obtain knowledge on ⇒ only suitable in MIMO
scenarios where is not needed
Types of self-calibration
Single broadcaster with multiple antennas [1-3]
Multiple broadcasters, each with one or more antennas [4-6]
7. 7Blue Mountain Wireless, Inc.
Drawbacks of Self-Calibration
Long calibration time
Calibration has to be in a serial fashion – antennas send
calibration signal one at a time to avoid interference
Calibration has to be performed for each transceiver gain setting
Each gain setting introduces different gain and phase offsets that
have to be calibrated
Each transceiver chain can have hundreds to thousands of gain
setting combinations (from various RF/analog stages)
Calibration also depends on carrier frequency
Serial processing for antenna, gain setting, and carrier results in
thousands to millions of combinations for calibration
Calibration process can take minutes or hours
Infeasible for massive MIMO
8. 8Blue Mountain Wireless, Inc.
Drawbacks of Self-Calibration
Calibration process, while long, also needs to be repeated
Gain/frequency-dependency changes over time
Drift and aging in analog components
Change in operational environment – temperature, antenna load, etc.
Calibration must repeat periodically
Repeated long calibrations reduce network capacity
More difficult for multiple broadcasters (antenna arrays in
different locations)
Multiple broadcasters run on independent oscillators – calibration
must include oscillator synchronization
GPS-based oscillator synchronization is expensive and requires
line-of-sight satellite paths
Over-the-air (non GPS-based) oscillator synchronization
Additional signaling protocol – further lengthening calibration process
More Tx power than single broadcaster calibration – creating
interferences in the network
9. 9Blue Mountain Wireless, Inc.
Drawbacks of Self-Calibration
Service disruption
Network service has to stop during calibration processes
Not acceptable in cellular networks where uninterrupted services
are expected
Self-calibration is not applicable in certain beamforming
scenarios
Recall that self-calibration only calibrates
Both and are needed in beamforming to multi-antenna
terminals
Difficult to standardize
Numerous self-calibration schemes exist, many of which are highly
device-dependent and include ad-hoc solutions
Ad-hoc nature hinders standardization thus industry-wide
acceptance and large-scale deployment
10. 10Blue Mountain Wireless, Inc.
Ramifications
There have been no feasible calibration methods for upcoming
5G and evolving 802.11 networks
Without efficient and accurate channel calibration, promise of
multi-folds capacity gain cannot be realized
Recall the unfulfilled promises by LTE CoMP
A feasible calibration method should ideally be
Fast – much shorter calibration time
Device-independent – no gain-dependent calibration
Accurate
Undisruptive to network services
Capable of calibrating , as well as if needed
Easy to standardize – accelerating acceptance and deployment
Generic – applicable to any wireless network, including 5G and
802.11
12. 12Blue Mountain Wireless, Inc.
Overview
Terminal-assisted calibration is an alternative to self-calibration
Broadcasters send downlink (DL) pilot signal to terminals
Terminals estimate DL channel and feed back to broadcasters
Terminals also send uplink (UL) pilot signal to broadcasters
Broadcasters estimate UL channel and derive and/or from UL
channel and feedback
Compared to self-calibration, terminal-assisted calibration has
been considered to be disadvantageous
Need to involve terminals
Large feedback overhead
We will describe a terminal-assisted calibration approach that
Overcomes drawbacks of self-calibration
Has extremely low feedback overhead
Fulfils desired properties of ideal calibrations
13. 13Blue Mountain Wireless, Inc.
Efficient Calibration Signal
Calibration signal is very efficient
Tones from all antennas can be mapped into one OFDM symbol
Antenna calibration function over the signal bandwidth can be
interpolated from calibration tones
Smoothness of calibration function ensures interpolation quality
Calibration signal can also be multi-symbol
Increases tone density thereby improving calibration accuracy
14. 14Blue Mountain Wireless, Inc.
Low Signaling and Feedback Overhead
Calibration overhead includes signaling and feedback1
Calibration signal can be as short as one OFDM symbol
An example for 20-MHz LTE with 256 antennas
There are 1200 subcarriers in one OFDM symbol
Each antenna can use four subcarriers for calibration
One Tx OFDM symbol suffices if long-term component2 is known
Types of feedback
Digital – about 5 symbols per Tx symbol per terminal antenna3
Analog – 2 ~ 4 symbols per Tx symbol per terminal antenna4
Thus calibration overhead can be well below 1 ms5
For multi-symbol calibration signal and multiple terminal antennas,
the overhead can still be on the order of milliseconds
For wider bandwidth, the overhead decreases accordingly
1 Low-overhead calibration tone design is based on principles described in US 8478203 and US 8792372
2 See later slides
3 Assumptions: 16 bits per (I,Q) symbol (8 bits for each of I and Q), and uplink spectral efficiency of 3.2 bits per subcarrier
4 Let = number of feedback symbols per Tx symbol. Assuming same SNR in uplink and downlink, = 1, 2, 4 correspond to
SNR degradation of 3.01, 1.76, 0.97 dB, respectively, due to analog feedback
5 For 20-MHz LTE, 1 ms = 14 OFDM symbols
15. 15Blue Mountain Wireless, Inc.
Simple Protocols
Terminal-assisted calibration follows very simple protocols
1. DL calibration signal is sent to participating terminals
Calibration signal consists of tones from all broadcaster antennas
2. Terminals send UL reference signals (RS)
For UL channel estimation
Existing UL RS in LTE/5G-NR can be used
3. Terminals feeds back received DL calibration tones
With UL channel estimation and DL calibration tone feedback,
broadcaster is able to derive and
16. 16Blue Mountain Wireless, Inc.
Gain Agnostic
Self-calibration process has to exhaust all gain settings
Calibration results are stored with respect to gain settings
The broadcaster chooses calibration result corresponding to
current gain setting
In terminal-assisted calibration, MIMO session immediately
follows calibration
Calibration always corresponds to most recent gain setting
Gain control is generally a low-rate process – one calibration can
apply to many subsequent MIMO sessions
A new calibration is only needed when gain setting is changed
17. 17Blue Mountain Wireless, Inc.
Improving Calibration Accuracy
In general, calibration quantities (dependency on frequency
added) have two components
Long-term component – fixed/slow-varying
Relative amplitude profiles
Relative nonlinear phases (there may also be none)
Relative antenna delays within same broadcaster
Short-term component – may change from time to time
Relative amplitude gains
Relative antenna phases
Relative antenna delays among broadcasters (e.g., in distributed
MIMO)
Short-term component consists of only scalar parameters,
instead of functions of frequencies as in long-term component
Assuming known long-term component, calibration turns into
parameter estimation
In general, estimating a few parameters has much higher accuracy
than reconstructing functions over the signal bandwidth
18. 18Blue Mountain Wireless, Inc.
Long-Term Component
Acquisition options for long-term component in
By initial calibration – longer than “normal” calibrations
Initial calibration can be designed to extract long-term component
with desired accuracy
By accumulating over normal calibrations
Avoid long initial calibrations
Calibration quality improves over time
Tracking and updating long-term component in
Updating information can be derived from normal calibrations to
track the slow variations in long-term component
What about ?
“Long-term” concept does not apply to terminals1 – each
calibration session may involve different terminals
can still be obtained accurately without long-term component
More broadcaster antennas than terminal antennas – e.g., 256 vs. 2
This translates to “processing gain” in estimating
1 “Long-term” concept is still relevant in terminal-centric calibrations – for example, in a “broadcaster-assisted calibration”
for a multi-antenna terminal
19. 19Blue Mountain Wireless, Inc.
More MIMO Modes
Terminal-assisted calibration is applicable to more MIMO modes
Modes where alone suffices
DL beamforming – one broadcaster, one single-antenna terminal
MU-MIMO – one broadcaster, multiple terminals
Distributed MIMO – multiple broadcasters, multiple terminals
Modes where both and are needed1
DL beamforming – one broadcaster, one multi-antenna terminal
MU-beamforming – one broadcaster, multiple multi-antenna terminals
Distributed MIMO/beamforming – multiple broadcasters, multiple multi-
antenna terminals
UL beamforming1: from a multi-antenna terminal to a broadcaster
Same calibration principle but with “role reversal” between the
broadcaster and the terminal
1 Self-calibration is not able to support these MIMO modes
20. 20Blue Mountain Wireless, Inc.
Flexible Terminal Selection
Terminal-assisted calibration has full flexibility in selecting
terminals
Calibration requires only one terminal antenna (for calibrating )
but can include more terminals or terminal antennas
Terminals in calibration are not necessarily the same as ones in
MIMO sessions
Benefits from terminal-selection flexibility
Choose terminals with best channel quality to maximize calibration
accuracy
Choose multiple terminals (or multiple terminal antennas),
including terminals not participating subsequent MIMO sessions
Multiple terminal antennas reduces impact of channel nulls, offering
effect of antenna diversity
Calibration accuracy is proportional to number of terminal antennas
21. 21Blue Mountain Wireless, Inc.
Ease of Standardization
Recall that it is difficult to standardize self-calibration
Self-calibration schemes depend highly on antenna-array properties and
many solutions are ad-hoc
Service disruption which is incompatible with cellular networks
In contrast, terminal-assisted calibration relies on generic principles
and operates on standard protocols
Calibration tones fit OFDM waveform in 4G, 5G-NR, and 802.11
DL/UL interaction fits in L1/L2 signaling of 3GPP
Well-known and standard signal processing algorithms
Calibration can be made a natural and integral part of MIMO operations
Wide adoption and deployment is possible only if calibration is
standardized
LTE-A CoMP fails to deliver multi-fold gain because of incomplete
standardization
Full CoMP is not feasible due to network constraints
Standardization barriers such as huge feedback overhead
Terminal-assisted calibration has none of the above issues
22. 22Blue Mountain Wireless, Inc.
WiFi is considered to play an indispensable role in 5G
WiFi networks are in many ways complementary to cellular
Operates in different bands from cellular
Small, self-organizing, and asynchronous
Wider bandwidth in sub-6 GHz: 160 MHz vs. 20 MHz in LTE
Free
WiFi is more prone to interferences and congestion
A single AP (access point) is ill-equipped serving large number of
terminals
Multiple APs interfere each other due to self-organizing and
asynchronous nature
WiFi throughput often comes to a standstill when networks are
dense and terminals are many
Massive MIMO can solve above WiFi issues and terminal-
assisted calibration enables it
802.11 Applications
23. 23Blue Mountain Wireless, Inc.
Coordinated AP transmission has been lacking in WiFi networks
Primary tool in reducing interference is interference avoidance
Interference avoidance in 802.11 has been primitive
Use difference channels in frequency domain
“Back-off” in time domain
Huge spectral inefficiency as a result
Distributed MIMO is more attractive than interference avoidance
802.11-based distributed MIMO was demonstrated in [3][4]
Self-calibration was used to restore channel reciprocity
Exhaustive calibration in gain space
prevents industry-wide adoption
Terminal-assisted calibration makes
distributed MIMO feasible
Phase-synchronizes APs
Eliminates gain-dependent calibration
802.11 – Distributed MIMO
24. 24Blue Mountain Wireless, Inc.
802.11 – Interference Avoidance
For APs with large antenna size, simultaneous MU-MIMO and
interference avoidance is possible
Each AP serves terminals in its own BSS and align its emission
nulls to neighboring APs and to terminals in other BSSs
Again, terminal-assisted calibration restores the needed channel
reciprocity
Simultaneous MU-MIMO and interference avoidance offer an
equally attractive alternative to distributed MIMO
No inter-AP backhaul is needed
No need for synchronizing APs – APs can operate in native
asynchronous WiFi mode
25. 25Blue Mountain Wireless, Inc.
Conclusion
The multi-fold gain of massive MIMO in TDD network can only be
realized by efficient channel calibrations
Terminal-assisted calibration possesses all desired properties
Fast – calibration time is on the order of milliseconds or less
Device-independent – no need to calibrate over the gain space
Accurate – separation of long-term and short-term components
improves calibration quality
Low signaling and feedback overhead
Capable of calibrating both broadcaster and terminal antennas –
supports more MIMO modes
Easy to standardize and undisruptive to network services –
calibration can be implemented with existing signaling protocols in
cellular networks
Applicable to both 5G and 802.11, and removing interference
bottleneck in 802.11
Terminal-assisted calibration is an enabling technology to TDD
massive MIMO in 5G
26. 26Blue Mountain Wireless, Inc.
References
[1] J. Vieira et al., “Reciprocity for massive MIMO: proposal, modeling, and validation”, IEEE
Trans. Wireless Comm., vol. 16, no. 5, pp. 3042–3056, May 2017.
[2] K. Gopala and D. Slock, “Optimal algorithms and CRB for reciprocity calibration in
massive MIMO”, IEEE International Conference on Acoustics, Speech and Signal
Processing, Calgary, Alberta, Canada, April 15-20, 2018.
[3] O. Raeesi et al., “Performance analysis of multi-user massive MIMO downlink under
channel non-reciprocity and imperfect CSI”, IEEE Trans. Comm., vol. 66, no. 6, pp. 2456–
2471, June 2018.
[4] US 9236998, “Transmitter and receiver calibration for obtaining the channel reciprocity for
time division duplex MIMO systems”, January 12, 2016.
[5] R. Rogalin et al., “Scalable synchronization and reciprocity calibration for distributed
multiuse MIMO”, IEEE Trans. Wireless Comm., vol. 13, no. 4, pp. 1815–1831, April 2014.
[6] H. Rahul et al., “MegaMIMO: scaling wireless capacity with user demands”, ACM
SIGCOMM 2012, Helsinki, Finland, August 2012.
[7] E. Hamed et al., “Real-time distributed MIMO systems”, Proceedings of the 2016 ACM
SIGCOMM conference, pp. 412-425, Florianopolis, Brazil, August 22–26, 2016.
[8] US 8792372, “Carrier-phase difference detection with mismatched transmitter and receiver
delays”, July 29, 2014.
[9] US 8478203, “Phase synchronization of base stations via mobile feedback in multipoint
broadcasting”, July 2, 2013.
[10] US 15/869042, “Phase synchronization and channel reciprocity calibration of antennas via
terminal feedback”, January 12, 2018.