Compression and decompression of downlink channel estimates

A method of operating a network node, the method comprising:

• determining a first set of parameters, the first set of parameters indicating a compression function for compressing downlink channel estimates at a user equipment (UE);
• determining a decompression function for decompressing downlink channel estimates which have been compressed by the UE using the compression function;
• transmitting the first set of parameters;
• receiving compressed downlink channel estimates;
• and
• decompressing the compressed downlink channel estimates using the decompression function,
• wherein the decompression function comprises a neural network having:
• a linear function;
• and
• a non-linear function,
• wherein determining the decompression function comprises:
• determining the linear function,
• wherein the first function is configured to:
• receive input data and
• provide output data in a higher dimensional space than the input data,
• wherein the non-linear function is configured to apply a non-linear activation function to each of a plurality of numbers in the output data of the linear function.

A user equipment (UE) comprising:

• processing circuitry;
• and at least one memory, the at least one memory containing instructions that when executed by the processing circuitry cause the UE to:
• receive a first set of parameters from a network node;
• form a compression function based on the first set of parameters;
• compress downlink channel estimates using the compression function;
• and
• transmit the compressed downlink channel estimates,
• wherein the compression function comprises a neural network having:
• a linear function;
• a non-linear function;
• and
• a quantizer,
• wherein the linear function is formed based on at least one of the parameters from the first set of parameters,
• wherein the linear function is configured to:
• receive input data;
• and
• provide output data with a lower dimension than the input data,
• wherein the non-linear function is configured to apply a non-linear activation function to each of a plurality of numbers in the output data of the linear function,
• wherein the quantizer is configured to:
• receive a plurality of numbers from the non-linear function;
• and
• apply scalar quantizers to the received numbers.

A network node comprising:

• processing circuitry;
• and
• at least one memory, the at least one memory containing instructions that when by the processing circuitry cause the network node to:
• determine a first set of parameters, the first set of parameters indicating a compression function for compressing downlink channel estimates at a user equipment (UE);
• determine a decompression function for decompressing downlink channel estimates which have been compressed by the UE using the compression function;
• transmit the first set of parameters;
• receive compressed downlink channel estimates;
• and
• decompress the compressed downlink channel estimates using the decompression function,
• wherein the decompression function comprises a neural network having:
• a linear function;
• and
• a non-linear function,
• wherein determining the decompression function comprises:
• determining the linear function,
• wherein the linear function is configured to:
• receive input data and
• provide output data in a higher dimensional space than the input data,
• wherein the non-linear function is configured to apply a non-linear activation function to each of a plurality of numbers in the output data of the linear function.