Protocol stack for analog communication in split architecture network for machine learning (ML) functions
A protocol stack architecture for processing machine learning (ML) data includes a ML layer to manage ML data communication with a network device. The ML layer is coupled to multiple ML training blocks, and ML and inference blocks for multiple neural networks, and an analog data communications stack coupled to the ML layer. The analog data communications stack has an upper media access control analog (MAC-A) layer coupled to the ML layer and configured to store data for each neural network, a lower MAC-A layer coupled to the upper MAC-A layer and configured to segment and reassemble analog ML data, and an analog physical layer coupled to the lower MAC-A layer and configured to communicate analog data with the network device. The architecture includes a digital data communications stack coupled to the ML layer and the lower MAC-A layer and configured to manage digital communications with the network device.
1. A protocol stack architecture for processing machine learning data at a user equipment (UE), comprising:
a machine learning layer configured to manage communication of machine learning data with a network device, the machine learning layer coupled to a plurality of machine learning training blocks, and machine learning and inference blocks for a plurality of neural networks;
an analog data communications stack coupled to the machine learning layer and comprising:
an upper media access control analog (MAC-A) layer coupled to the machine learning layer and configured to store analog data for each of the plurality of neural networks;
a lower MAC-A layer coupled to the upper MAC-A layer and configured to segment and reassemble the analog data; and
an analog physical layer coupled to the lower MAC-A layer and configured to transmit and receive the analog data to and from the network device; and
a digital data communications stack coupled to the machine learning layer and the lower MAC-A layer and configured to manage digital communications with the network device.
2. The architecture of claim 1 , in which the upper MAC-A layer is further configured for analog ciphering of the analog data, bearer management for the analog data, and packet management for the analog data.
3. The architecture of claim 1 , in which the lower MAC-A layer is further configured to map the analog data to component carriers, to manage data retransmission for the analog data, and to perform multiplexing and de-multiplexing of the analog data from the plurality of neural networks.
4. The architecture of claim 1 , in which the analog physical layer is further configured to control a bandwidth size for the analog data, to map to a transmit waveform, to de-map from a receive waveform, to map the analog data to transmit ports, and to manage beamforming, pre-equalization and phase coherence.
5. The architecture of claim 1 , in which the lower MAC-A layer is further configured to receive control messages from a MAC layer of the digital data communications stack.
6. The architecture of claim 5 , in which the control messages dynamically configure the analog data communication stack.
7. The architecture of claim 1 , in which functions of the analog data communication stack are performed by the digital data communications stack.
8. A method of wireless communication, by a user equipment (UE), comprising:
receiving a control message from a machine learning training block or machine learning inference block;
transmitting, to a network device, the control message via a digital data communications stack;
receiving gradient data for federated learning, from the machine learning training block or the machine learning inference block;
determining whether to transmit the gradient data via an analog data communications stack or the digital data communications stack based on a network configuration; and
transmitting, to the network device, the gradient data via the analog data communications stack or the digital data communications stack in accordance with the determining.
9. The method of claim 8 , further comprising receiving the network configuration dynamically.
10. The method of claim 9 , further comprising receiving the network configuration via radio resource control signaling.
11. The method of claim 8 , in which the network configuration is pre-configured.
12. The method of claim 8 , in which the determining is performed packet-by-packet.
13. The method of claim 8 , in which the determining is performed stream-by-stream.
14. The method of claim 8 , in which the network configuration is based on a quantity of UEs participating in the federated learning.
15. The method of claim 8 , in which the transmitting is via an analog physical uplink shared channel (PUSCH-A).
16. The method of claim 8 , further comprising receiving, from the network device, neural network weights and/or the gradient data via an analog physical downlink shared channel (PDSCH-A).
17. The method of claim 16 , further comprising receiving, from the network device, the neural network weights and/or the gradient data via a broadcast channel.
18. The method of claim 8 , in which transmitting via the analog data communications stack comprises communicating with the digital data communications stack with an indication that analog functions should be performed.
19. A protocol stack architecture for processing machine learning data at a network device, comprising:
a machine learning layer configured to manage communication of machine learning data with a user equipment (UE), the machine learning layer coupled to a plurality of machine learning training blocks, and machine learning and inference blocks for a plurality of neural networks;
an analog data communications stack coupled to the machine learning layer and comprising:
an upper media access control analog (MAC-A) layer coupled to the machine learning layer and configured to store analog data for each of the plurality of neural networks;
a lower MAC-A layer coupled to the upper MAC-A layer and configured to segment and reassemble the analog data; and
an analog physical layer coupled to the lower MAC-A layer and configured to transmit and receive the analog data to and from the network device; and
a digital data communications stack coupled to the machine learning layer and the lower MAC-A layer and configured to manage digital communications with the UE.
20. The architecture of claim 19 , in which the upper MAC-A layer is further configured for analog ciphering of the analog data, bearer management for the analog data, and packet management for the analog data.
21. The architecture of claim 19 , in which the lower MAC-A layer is further configured to map the analog data to component carriers, to manage data retransmission for the analog data, and to perform multiplexing and de-multiplexing of the analog data from the plurality of neural networks.
22. The architecture of claim 19 , in which the analog physical layer is further configured to control a bandwidth size for the analog data, to map to a transmit waveform, to de-map from a receive waveform, to map the analog data to transmit ports, and to manage beamforming, pre-equalization and phase coherence.
23. The architecture of claim 19 , in which the lower MAC-A layer is further configured to receive control messages from a MAC layer of the digital data communications stack.
24. The architecture of claim 23 , in which the control messages dynamically configure the analog data communication stack.
25. The architecture of claim 19 , in which functions of the analog data communication stack are performed by the digital data communications stack.
26. A method of wireless communication, by a network device, comprising:
receiving a control message from a machine learning training block or machine learning inference block;
transmitting the control message via a digital data communications stack;
receiving gradient data for federated learning, from the machine learning training block or the machine learning inference block;
determining whether to transmit the gradient data via an analog data communications stack or the digital data communications stack based on a network configuration; and
transmitting the gradient data via the analog data communications stack or the digital data communications stack in accordance with the determining.
27. The method of claim 26 , further comprising transmitting neural network weights and/or the gradient data via an analog physical downlink shared channel (PDSCH-A).
28. The method of claim 27 , further comprising transmitting the neural network weights and/or the gradient data via a broadcast channel.
29. The method of claim 26 , in which the network configuration is pre-configured.
30. The method of claim 26 , in which transmitting via the analog data communications stack comprises communicating with the digital data communications stack with an indication that analog functions should be performed.