Latency reduction based on packet error prediction

A method for transmitting data packets over one or more wireless communication links to one or more receivers, the method comprising:

• maintaining, in a database, a trained machine learning algorithm for predicting a probability of failed delivery of data packets over the one or more wireless communication links based on measurements relating to the one or more wireless communication links;
• receiving a raw measurement data stream by performing measurements over the one or more wireless communication links;
• separating, from the raw measurement data stream, one or more data streams associated with the one or more wireless communication links;
• if any of the one or more data streams comprises string data, performing word embedding for said string data,
• filling any temporal data gaps in the one or more data streams with surrogate values to form one or more pre-processed data streams;
• performing feature engineering on the one or more pre-processed data streams to form a measurement data set for each of the one or more wireless communication links;
• and
• performing, for each of the one or more wireless communication links as a first wireless communication link, the following:
• predicting, using the trained machine learning algorithm, a probability of failed delivery of a data packet over the first wireless communication link to a first receiver, wherein a corresponding measurement data set is used as an input of the trained machine learning algorithm,
• determining whether the probability of failed delivery of the data packet over the first wireless communication link exceeds a threshold, and
• in response to the probability of failed delivery of the data packet exceeding the threshold, transmitting the data packet over the first wireless communication link to the first receiver and transmitting to the first receiver one of a copy of the data packet and redundancy data for the data packet.

A computer program embodied on a non-transitory computer-readable medium, said computer-readable medium including instructions stored thereon which, when run on hardware, control the hardware to perform at least the following:

• receiving a raw measurement data stream by performing measurements over the one or more wireless communication links;
• separating, from the raw measurement data stream, one or more data streams associated with the one or more wireless communication links;
• if any of the one or more data streams comprises string data, performing word embedding for said string data,
• filling any temporal data gaps in the one or more data streams with surrogate values to form one or more pre-processed data streams;
• and
• performing feature engineering on the one or more pre-processed data streams to form a measurement data set for each of the one or more wireless communication links;
• performing, for each of the one or more wireless communication links as a first wireless communication link, the following:
• predicting, using a trained machine learning algorithm with a corresponding measurement data set as an input, a probability of failed delivery of a data packet over the first wireless communication link to a first receiver, wherein the trained machine learning algorithm is a trained algorithm for predicting a probability of failed delivery of data packets over the one or more wireless communication links based on measurements relating to the one or more wireless communication links,
• determining whether the probability of failed delivery of the data packet over the first wireless communication link exceeds a threshold, and
• in response to the probability of failed delivery of the data packet exceeding the threshold, transmitting the data packet over the first wireless communication link to the first receiver and transmitting to the first receiver one of a copy of the data packet and redundancy data for the data packet.