Learning approximate estimation networks for communication channel state information

A system comprising:

• at least one processor;
• and
• at least one computer memory coupled to the at least one processor having stored thereon instructions which, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
• processing first information with ground truth information to generate a first radio-frequency (RF) signal at least partially by altering the first information by a channel impairment having at least one channel effect;
• using at least one receiver to process the first RF signal to generate second information;
• training a machine-learning estimation network based on a network architecture, the second information as an estimator input, and one or more items of the ground truth information as one or more estimator target outputs to generate learned estimation information;
• receiving, using the at least one receiver, a second RF signal transmitted through a communication channel comprising the at least one channel effect;
• inferring, by the trained machine-learning estimation network, the receiver to estimate an offset of the second RF signal caused by the at least one channel effect;
• and
• correcting the offset of the RF signal with the estimated offset to obtain a recovered RF signal.

A non-transitory computer-readable storage medium coupled to at least one processor having machine-executable instructions stored thereon that, when executed by the at least one processor, cause the at least one processor to perform operations for deploying one or more machine-learning estimation network to a communications system, the operations comprising:

• receiving, by at least one receiver, a RF signal transmitted through a communication channel having at least one channel effect;
• inferring, by a previously-trained machine-learning estimation network, the receiver to estimate an offset of the RF signal caused by the at least one channel effect, the machine-learning estimation network being trained based on the at least one channel effect;
• and
• correcting the offset of the RF signal with the estimated offset to obtain a recovered RF signal.