the road towards an ai-native air interface (ai-ai) for 6g
TRANSCRIPT
1
The Road Towards an AI-Native Air Interface (AI-AI) for 6G
Jakob HoydisRadio Systems Research & AI
Nokia Bell Labs, France
2/28
Creating the โaugmented humanโ
What will future communications look like in 2030?
6G to enable a new lifestyle at scale
Learn from/with machines
Automatic security
In-body monitoring
Augment our Intelligence
Augment our experience
High resolution mapping
Mixed reality co-design
Multi-model mixed reality telepresence
Domestic robots
Remote & self driving
Drone/robot swarms
Augment our control
3/28
Human Machine InterfacePrecision Sensing & Actuation
Knowledge systems
Biological World
Digital World
6G
Physical World
Real time
Ubiquitous Compute
The enabling foundation for that futureโฆ
H. Viswanathan and P. Mogensen, โCommunications in the 6G Eraโ, IEEE Access, 2020
4/28
Six key technologies for 6G
AI/ML Air-Interface
Extreme Connectivity
Network as a sensor
Security and Trust
New Spectrum Technologies
RAN-Core Convergence & Specialization
5/28
Random Access detection Symbol demapping MIMO detectionChannel estimation MIMO User Pairing
Deployment optimizationLoad balancingCarrier Aggregation Handover predictionBeam PredictionInterference Management
Slice orchestrationVM managementAnomaly detection
User localization
Trajectory prediction
No component of 5G has been designed by ML
DU
AMF UPF
SMF CU
Lack of accurate modelsAlgorithms not optimalComplexity
Role of ML for 5G
6/28
What if 6G was built so that ML could optimizeparts of the PHY & MAC if needed?
7/28
Possible benefits
Optimally adapted to hardware limitations
No heuristic parameter settings
Faster development & deployment time
Bespoke waveforms, constellations & pilots
Reduced standardization effort
Custom signaling and access schemes
AI-AI
8/28
AI-Native Air Interface (AI-AI) for 6G
HardwareData
Environment, hardware & application dependency feedback
Hardware ApplicationEnvironmentAI-AI AI-AI
The AI-AI optimally adapts to different environments, hardware, data, and applications
โPost Shannonโ: Not about reliably transmitting bits anymore, but rather serving an application with data in an optimal way
9/28
A roadmap to an AI-Native Air Interface for 6G
EncodingSymbol Mapping
Modulation / Waveform
Transmitter
Sync.Channel
EstimationEqualization
Symbol Demapping
Decoding
Receiver
ML ML ML ML ML1. ML replaces/enhances individual processing blocks
ML ML ML2. ML replaces multiple processing blocks
ML ML
3. ML designs part of the PHY itself
6G
10/28
The importance of each step in the transition
3. ML designs part of the PHY itself
โข Paradigm change in communication systems design
โข Distributed & end-to-end learning
โข New signaling & procedures
1. ML replaces/enhances individual processing blocks
โข Paradigm change in transceiver design & deployment
โข Online training & transfer/federated learning
โข No new signaling
2. ML replaces multiple processing blocks
โข Enables qualitative new features (e.g., no cyclic prefix, less pilots)
โข HW acceleration essential
โข ML-first approach without backup solution
Sync.Channel
EstimationSymbol
demappingDecoding
Receiver
Sync. Decoding
Receiver
Transmitter Receiver
11/28
Differentiable algorithms enable hybrid โML-Expertโ systems
Training on end-to-end loss :
โข No need for block-wise ground-truth
โข Each block optimized for end-to-end performance
Sync.Channel
EstimationEqualization
Symbol Demapping
Decoding
Receiver
โ ๐๐
๐๐๐ ๐น
๐
ร ๐
๐๐
Fully differentiable transceivers allows simple integration of trainable components
12/28
A universal end-to-end design goal
Receiver
๐๐๐
Channel๐(๐ฆ|๐ฅ)
Transmitter
๐๐๐
๐ฅ ๐ฆ๐(๐1|๐ฆ)
โฎ๐(๐๐|๐ฆ)
๐1โฎ๐๐
โ ๐๐ , ๐๐ =
๐=1
๐
โ๐ผ๐๐,๐ log2 ๐๐ ๐๐ ๐ฆ
Bit-metric decoding rate
(depends on transmitter)
Loss due to imperfect receiver
(depends on receiver)
Number of transmitted bits
Achievable rate with a mismatched bit-metric decoder
Binary cross-entropy
Minimizing the binary cross-entropy maximizes an achievable rate with a practical decoder
= ๐ โ ( ฯ๐=1๐ ๐ผ ๐๐; ๐ฆ โ ฯ๐=1
๐ ๐ผ๐ DKL ิก๐ ๐๐ ๐ฆ ๐ ๐๐ ๐ฆ )
Estimated posteriorbit probabilities
13/28
Understanding end-to-end learning on an AWGN channel
Curtesy of Sebastian Dรถrner, University of StuttgartS. Cammerer, et al. "Trainable communication systems: Concepts and prototype," IEEE Trans. Commun., no. 9, vol. 68, Sep. 2020.
Constellation determinesbit-metric
decoding rate
Decision regions
determine loss w.r.t. MAP
14/28
Case study:
From Neural Receivers to Pilotless Transmissions
15/28
SISO doubly-selective channel
Spectral correlation
โข ๐น๐น ๐,๐ = ฯ๐=1๐ฟ ๐๐๐
๐2๐๐๐๐ท๐ ฮ๐น(๐โ๐)
โข Subcarrier spacing ฮ๐น = 30 kHzโข Delay spread ๐ท๐ = 100 nsโข TDL-A power delay profile
Temporal correlation
โข ๐น๐ ๐,๐ = ๐ฝ0 2๐๐ฃ
๐๐๐ฮ๐(๐ โ ๐)
โข Carrier frequency ๐๐ = 3.5 GHzโข Speed ๐ฃ = 50 km/h14 OFDM symbols
72 Subcarriers
pilots
๐ = ๐ฏ โ ๐ฟ + ๐ต
vec ๐ฏ โผ ๐๐ฉ(0,๐น๐นโจ๐น๐)
Resourceelement (RE)
Modulation & Coding
โข 64 QAMโข 5G code n=1024, r=2/3
16/28
Baseline receiver
โข Least-squares channel estimation at pilot positions
โข Equalization using the nearest pilot
โข Exact LLR computation assuming a Gaussian post-equalized channel
โข Textbook sum-product BP decoder with 40 iterations
LS estimate
REs equalized using the nearest pilot
Channel Estimation
EqualizationSymbol
demappingDecoding
Receiver
17/28
Potential performance enhancements
18/28
Deficits of the baseline
Imperfect channel estimation & channel aging lead to
โข Mismatched LLR computation
โข SNR degradation
Su
bca
rrie
rsOFDM symbols
19/28
Neural demapper (symbol-wise)
๐ ๐ ๐๐,๐
๐ผ๐ ๐๐,๐
Pos. enc. ๐
Pos. enc. ๐
๐๐๐ ๐,๐
๐ฟ๐ฟ๐ ๐,๐(1)
๐ฟ๐ฟ๐ ๐,๐(6)
โฎโฎ
Channel Estimation
Equalization Decoding
Receiver
Learns grid position-dependent statistics for better LLR computation
20/28
Neural demapper (grid-wise)
Channel Estimation
Equalization Decoding
Receiver
๐ ๐ ๐ฟ , ๐ผ๐ ๐ฟ , ๐๐๐
72 ร 14 ร 3
๐ณ๐ณ๐น
72 ร 14 ร 6
Leverages pilots and data to compensate for channel aging and mismatched LLR computation
21/28
Neural network architecture is key to success
H. Talebi et al., โLearned perceptual image enhancementโ, arXiv:1712.02864F. Yu et al., โMulti-scale context aggregation by dilated convolutionsโ, arXiv:1511.07122
https://medium.com/@zurister/depth-wise-convolution-and-depth-wise-separable-convolution-37346565d4ec
Each output value has a receptive field spanning the entire resource grid
M. Honkala, et al., โDeepRX: Fully ConvolutionalDeep Learning Receiverโ, arXiv2005.01494
Fully convolutional ResNet Dilated separable convolutions
22/28
Neural receiver
Decoding
Receiver
๐ ๐ ๐ , ๐ผ๐ ๐
72 ร 14 ร 2
๐ณ๐ณ๐น
72 ร 14 ร 6
Data-aided channel estimation, equalization, and demapping for unprecedented performance
23/28
End-to-end learning with Neural receiver
End-to-end learning enables pilotless transmissions without performance loss
EncodingModulation /
Waveform
Transmitter
๐ =แ๐ โ
164
ฯ๐โ แ๐ ๐
164
ฯ๐โ แ๐
๐ 2 โ164
ฯ๐โ แ๐
๐2
Zero-mean unit energy constellation
64x2 trainable weights
24/28
Learned constellation for pilotless communication
How could this be standardized?
F. Ait Aoudia, J. Hoydis, โEnd-to-end Learning for OFDM: From Neural Receivers to Pilotless Communicationโ, arXiv2009.05261
25/28
Important research topics for end-to-end learning
โข New waveforms for new spectrum
โข Learning for systems with (extreme) hardware constraints
โข Joined communications + X
โข Signals conveying a few bits of information
โข Application-specific end-to-end learning
โข Semantic communications
โข Decentralized & federated learning
โข Transfer & meta learning
26/28
The next frontier:
Protocol learning
27/28
Emerging a RAN protocol
A. Valcarce and J. Hoydis, โTowards joint learning of optimal MAC signaling and wireless access channel accessโ, arXiv:2007.09948v2
3GPP Way
vs RF
ML engine
ML Way
Can we learn a MAC protocol?
28
Thank you!