Researchers at MIT, Maynooth University and Boston University have recently created the first silicon chip using a universal decoding algorithm called GRAND aka Guessing Random Additive Noise Decoding, which can decode any code more accurately, regardless of its structure. .
GRAND has eliminated the need for multiple computationally complex decoders. The chip has allowed for greater efficiency that could have augmented and virtual reality (AR / VR) applications, 5G networks, games and connected devices that process a large volume of data with minimal delay.
Deep Learning DevCon 2021 | 23-24th Sep | Register>>
With the support of the Battelle Memorial Institute and the Science Foundation of Ireland, the research is expected to be presented at the European Conference on Solid State Devices and Circuits (ESSCIRC ESSDERC) held this week.
How noise makes data transfer difficult
All information traveling over the Internet, from paragraphs in an email to 3D graphics in a virtual reality environment, can be altered by the noise along the way, for example, by the electromagnetic interference of a microwave or Bluetooth device. Typically, the data is encoded so that, when it reaches its destination, a decoding algorithm can undo the negative effects of these perturbations and recover the original data.
Traditionally, most error correction decoding codes and algorithms have been designed together. As a result, each code had a structure that corresponded to a very complex decoding algorithm, which often required dedicated hardware. Thanks to GRAND, it has eliminated the need for multiple and complex hardware components.
Looking for a job change? Let us help you.
To understand how GRAND works, we think of these codes as redundant hashes (1s and 0s) added to the end of the original data. The rules for creating this hash are stored in a specific codebook.
As encoded data travels over the network, it is affected by signal-altering noise, which is often generated by electronic devices. Therefore, when they arrive (encoded data and the noise that affected them) at their respective destination, the decoding algorithm checks their codebook and uses the hash structure to guess what the stored information is.
The way GRAND works is that it guesses the noise that affects the message and uses the noise pattern to deduce the original information. It generates a series of noise sequences in the order in which they are likely to occur, deduces them from the received data, and checks to see if the code word is in a codebook.
This is made possible because noise has a particular structure that allows the algorithm to guess what it may be, even if the noise appears random. Muriel Médard, an MIT researcher, said it is similar to problem solving.
Giving an example of a mechanic shop, he said, “If someone takes their car to the store, the mechanic doesn’t start by mapping the whole car to plans. Instead, they start by asking: what is most likely to go wrong? Maybe you just need gas. If that doesn’t work, what will happen next? Maybe the battery has run out. ”
Inside the GRAND chip
The GRAND chip uses a three-tier structure, which includes the simplest possible solutions in the first stage and longer and more complex noise patterns in the later two stages. As a result, each stage works separately, which increases system performance and saves energy.
In addition, the device is designed to switch seamlessly between two codebooks: one breaks the code words, while the other loads a new codebook and then goes into decoding without downtime or delay. .
As for the experimental result, the researchers found that the GRAND chip could effectively decode any redundancy code up to 128 bits in length, with only about a microsecond of latency.
Earlier, MIT researchers had demonstrated the success of the algorithm, but with their latest work, they have been able to show the effectiveness and efficiency of GRAND in hardware. Médard said the development of hardware for the new decoding algorithm required researchers to first set aside their previous concepts.
He said they could have gone out and reused things that were already being done. But they decided to rethink all aspects from scratch. “It was a journey of reconsideration,” Médard added.
Because GRAND uses codebooks for verification, researchers believe the chip not only works with inherited codes, but can be used with codes that have not even been entered.
For example, in the case of 5G implementation, telecom providers and regulators have difficulty finding common ground for the codes to be used in the new network. Unfortunately, regulators often choose traditional codes for 5G infrastructure in different scenarios. Taking advantage of GRAND could help eliminate the need for rigid standardization in the future, Médard said.
In addition, researchers believe their chip could even usher in a new wave of coding innovation. “I hope this will reshape the discussion, so that it is not standards-oriented, which will allow people to use codes that already exist and create new codes,” he added.
In the coming months, researchers plan to address issues related to software detection with a new version of the GRAND chip, as the data received in program detection is less accurate. In addition, they plan to test the chip’s ability to break longer and more complex codes and further adjust the structure of the silicon chip, improving its energy efficiency.
Join our Discord server. Be part of an attractive online community. Join here.
Subscribe to our newsletter
Share your email with the latest updates and relevant offers.