We have now released the 31st episode of the podcast Wireless Future. It has the following abstract:
A wave of digitalization is sweeping over the world, but not everything benefits from a transformation from analog to digital methods. In this episode, Emil Björnson and Erik G. Larsson discuss the fundamentals of analog modulation techniques to pinpoint their key advantages. Particular attention is given to how analog modulation enables over-the-air aggregation of data, which can be used for computations, efficient federated training of machine learning models, and distributed hypothesis testing. The conversation covers the need for coherent operation and power control and outlines the challenges that researchers are now facing when extending the methods to multi-antenna systems. Towards the end, the following paper is mentioned: “Optimal MIMO Combining for Blind Federated Edge Learning with Gradient Sparsification”.
You can watch the video podcast on YouTube:
You can listen to the audio-only podcast at the following places:
Dear Emil,
Many thanks for the useful podcast. In analog OTA, a denoising or scaling factor is applied at the server to mitigate the effects of aggregation error. It would be highly appreciated if you could explain whether it is applied in the analog domain or digital domain.
Many thanks.
I think it can be done in either of these domains, but it would be easiest to do it digitally.
In particular, if one does analog OTA over many subcarriers, then a digital implementation could apply different scaling factors at different subcarriers.
Respected Professor Emil,
Thanks for your insights on OTA.
In over-the-air aggregation (OTA), we receive superimposed signals. By assuming if 10 nodes transmitted their data, at the receiver, is it possible to recover each node’s data individually along with their sum or average? What can be a possible approach?
Usually in OTA, it’s being said that all nodes are contributing equally. What if they don’t contribute equally. We have 10 nodes, and only 6 are participating at some instance. How can we make sure that we are getting the average across those 6 devices and not from all 10 devices?
This is a good question. In a normal system, the nodes transmit their data separately to the base station, which then gain access to the individual data and can later compute any function of them (e.g., sum or average). If we have 10 nodes, then we need to create 10 parallel channels for this, through OFDMA (4G) or MU-MIMO (5G).
The idea with OTA aggregation is that we are not interested in the individual data, but only the sum of it. We can then only need one channel for this, where the base station can only receive the summation.
Regarding the 6 vs. 10 nodes: The receiver will get the sum of the data. When calculating the average, it needs to divide with the number of participating nodes. I think we can assume that this number is known to the base station since it anyway needs to instruct the nodes on what power values and phase-shifts to use when transmitting, to compensate for their different complex-valued channel gains.