Marginal gains and how they impact latency.

‘Marginal gains’ is a term that has come into common sporting parlance, if not everyday life. Back in 2003, when England won the Rugby World Cup in Sydney, Sir Clive Woodward attributed it to marginal gains - doing 100 things 1% better, rather than one thing 100% better. One ‘1%’ was England’s skin-tight shirts. Back in the age of ‘baggy’ rugby shirts, this was a revelation, meaning a player couldn’t be (easily) tackled by grabbing the shirt, as happened in an earlier England game, costing them the match. Roll forwards 5 years, all professional teams had them…

Dave Brailsford, the GB Cycling Team Manager / Performance Director 1997-2014 and Team Sky’s Manager, is credited with championing a philosophy of ‘marginal gains’ at British Cycling. Many travelers take their own pillow with them when they travel, GB Cycling provides their team with thin memory foam mattresses to guarantee a good night’s sleep in a strange bed!

Looking at the leap in performance needs between 4G (LTE) and 5G, many might feel this an insurmountable problem, but if the problem is broken down, we can look for those marginal gains.

LTE-A Pro specifies a maximum throughput of 2Gbps with an end-to-end latency of 1ms. 5G has an aggressive spec of 20Gbps with an end-to-end latency of 0.5ms

Now surprisingly, the throughput increase is not too difficult to overcome, (see Prof Maunder’s video on our website,, simply putting channel coding blocks in parallel will increase the rate, albeit with an increase in silicon area. The latency however is a bigger problem, the same trick doesn’t work, you’re still restricted by the same architecture.  

End-to-end, as the name implies, is measured from the signal leaving the basestation to when it is received at the handset (or vice versa). In 2014 Professor Fettweis of TU Dresden calculated that 50% of the latency was taken up in the radio signal chain, the rest being over the air transmission time.

Some of the processing in the signal chain is fixed, based on the throughput, however, if you could find a way to reduce the latency through the channel coder, some latency could be saved. AccelerComm managed to do this with their turbo decoder, it has 10X lower latency than any turbo decoder commercially available. It reduced the worst-case decoding latency from 50µs to 5µs on a mid-range FPGA - a saving of 45µs. It was designed with 5G NR in mind, but sadly 3GPP chose LDPC for the data channel. For the low TTI needed in LTE-A Pro for LTE URLLC it is ideal, but an order of magnitude might not be considered a ‘marginal gain’!

AccelerComm have used similar techniques in their 5G NR polar IP for the control channel, again providing lower latency implementations than anything published to-date. Based on comments at MWC18 in Barcelona last week, we’ve only gone and done it again!

Let’s look at what impact this marginal gain can really have, after all 45µs in 500µs is ‘nothing’…isn’t it?

Poor channel coding in the data channel will result in a reduced data throughput (based on the need for retransmissions etc), a poor control channel is a different issue, retransmission is only part of it. Firstly, your data channel isn’t set up at all, so no data. Second, system performance is degraded, you’ll have poor QoS, dropped calls and some very dissatisfied customers.

Saving ‘a few’ µs of latency in the channel coder could make a huge difference to the overall performance of the channel, as well as the challenge of hitting the end-to-end latency budget. That delay you get when the plane lands and you’re waiting for your phone to ‘find a signal’, that can be improved …a marginal gain for all travelers.

It would appear that marginal gains in the control channel can amount to significant benefits to operators (happy customers through reliable service) to OEMs (through reduced retransmissions, saving energy and compute time) and to us as users, getting connected faster at our destination.

Channel coding isn’t a tick box, one size fits all anymore, 5G NR needs experts to deliver the best systems and networks…DIY is one acronym that doesn’t belong in 5G!

Organisations we work with