Tony Poulos, TelecomTV (00:07):
Hi, I am Tony Pulos and I'm at Mobile World Congress in Barcelona. And today I have with me Harpinder Matharu, who is the senior director product marketing, the Wired and wireless group at AMD. Harpinder, It's great to catch up. I'm really interested to hear about what you are doing here, particularly around AI because everyone's talking about AI and how does AI fit into the RAN and the open ran market?

Harpinder Matharu, AMD (00:32):
Yeah, so AI should look at it as a tool in a toolbox, and there are three aspects. First of all, within the base station you want to find avenues to improve the capacity of the base station. So AI comes very handy there wherever we have estimation. Channel estimation is one aspect of it. Wherever you have heuristic algorithms like scheduler, AI can do, marvelous. And when you look at a cluster of base stations, how you manage them, AI is a great tool for that. And then if you go to Edge or where you host your services, that is another area where AI tends to be a very good tool. So from that perspective, ran and open RAN is trying to define some use cases in context of Rick and I think they're going to expand it to other areas of RAN as well. So we are looking forward to its use cases and three, GPP is now defining it as part of their release 17. So it's getting covered at all aspects of standardization and industry lines as well as we are seeing that these tools can be leveraged today in context of 5G and then leading into six G.

Tony Poulos, TelecomTV (01:42):
So would I be right in presuming that the AI could help produce a more efficient RAN system and also, but is there any power or performance degradation? We keep hearing about these GPUs being really power hungry,

Harpinder Matharu, AMD (01:55):
Right, so I think a lot of things need to be clarified here. First of all, in a base instance you don't need A GPU and it's basically you can instantiate AI as an inference engine as part of, say for example an A MD is doing adaptive SOC products and we have a demo going on here where we use a portion of our device to do AI inference. And just to give an example, one of the key requirement that we see today is how do we really improve the spectral efficiency, for example. So in order to do that, we are using portion of our device which sits in existing today's base station as a companion device to sometimes Asics. So the same device can be leveraged to boost the spectral efficiency. And how you do that, you do a better channel estimation. Everything is having some pattern to it.

(02:44)
We as a user have a pattern to using the network. So you leverage those patterns and improve the channel estimation in that process. And thereby you can enable the scheduler to schedule users more efficiently. So if you're able to the spectral utilization by say 20 or 30%, operators can save the number of base station instance they need to have or they can scale the capacity right now when it comes to question of power. So these devices are designed for tops per dollar, per wat efficiency. So they're extremely efficient from that perspective. Now having said that, certainly you need GPUs for training and that training you have to do as part of your model development. But once you train your models, you instantiate these models in these devices and you go through a lot of optimization techniques like you do quantization, you do pruning so that these models are efficiently implemented and don't cause penalty. In fact the ROI will be much better. So we are working on these exciting technologies and they'll mature as the time progresses.

Tony Poulos, TelecomTV (03:54):
I dread to ask this question, but can you elaborate more on AI and six G and how do those two correlate?

Harpinder Matharu, AMD (04:02):
Wow, it's a very good question. In fact, as you know, six G requirements are being gathered in three gpp standardization will happen. But that three GPP was already putting AI as a very important element in the specifications. So we believe that just like 5G introduced the cloud native concept, we believe that six G 3G PP and the proposal that we will see in three GP, we will move it towards being AI native. So leveraging AI as a tool and on all aspects of RAN deployment in the signal chain within the base station to how you optimize it to how you instantiate services. We do believe that there'll be a core part of 60. And another important thing to highlight here is for 5G, for example, release 15 started with enhanced mobile broadband as a main use case targeting consumers. And then subsequently we have other use cases coming into picture like ultra reliable, low latency and massive machine type communication. We believe that six G needs to look at enterprise and other services first so that we can scale and we can enable help operators to monetize their networks so that they can pass on that benefits to the supply chain. So the supply chain could be a little bit more being productive, beaming with lot more innovations coming in. So we believe that 60 will be more enterprise centric and will open up more use cases for the RAN deployments.

Tony Poulos, TelecomTV (05:45):
What are the challenges that AMD is seeing on the RAN and ai?

Harpinder Matharu, AMD (05:49):
Right, if you look at AI concept is first of all, some of the AI elements have already been introduced, have been implemented an existing base station implementation. So they're not something completely new. But if you have to incorporate some of the new developments that we have seen in ai, obviously we have to find a way to introduce them in the existing infrastructure. So I would say that a lot of radios are already existing, they're installed. The first place where we can introduce them is the DU or the base station and as we're seeing V ran instances are getting introduced. So it's much easier for us to introduce AI in a v ran instance in ddu. And obviously as newer base sessions evolve, they'll be much more planned. If it is a proprietary box, they'll have some functionality already incorporated. So we do see that beginning will be DU and gradually as new radios are deployed and you introducing those radios and likewise you provision those in the edge sites as well.

Tony Poulos, TelecomTV (06:54):
In closing, 6G is stated to be deployed in 2030. So why are we talking about it now?

Harpinder Matharu, AMD (07:02):
Yeah, first I would like to correct that we look forward to seeing some early deployments in 2028 as part of has been the case. Olympics will trigger that it'll be early deployments, but if they're early deployments, think 2028, you need to have systems ready by 26, 27 tested. Okay, there'll be standardization, may not be quite there, but still you have to have some systems tried out and if you have to do standardization then you need to have some proof of concept. Some test beds in 20 26, 20 25. So we as adaptive S-O-C-F-P-G-A provider play a significant role right from the beginning. When you have an idea, you do a proof of concept to prove it out and then that becomes one of the contributions to the standard. And then you do the take the proof of concept for early commercialization, early introduction of the products and then subsequently mainstream. So you have to basically look from 2028 backwards and therefore I think it's about time we are in 2024, so not much time remaining for us and

Tony Poulos, TelecomTV (08:06):
We've still got a lot of 5G catching up to do in the meantime.

Harpinder Matharu, AMD (08:10):
I agree. I think 5G still more deployments need to come and we do see that many of the sites, they are going to hit capacity given every year we have a growth of 30% more traffic. So I think 60, just like 5G introduced these mid bands. We expect that 60 will open up seven gigahertz to 24 gigahertz. Again, which specific band that has to be decided as within the industry. But so for these bands it's potentially possible that we'll look into new transmission schemes, new frame structure, and new coding. Right. So from that perspective we do see that 60 G discussion will start in parallel replenishing the 5G infrastructure.

Tony Poulos, TelecomTV (08:57):
We will never stop moving forward. Harpinder, thank you very much for being with me today.

Harpinder Matharu, AMD (09:01):
Thank you Tony.