What's Up with Tech?
Tech Transformation with Evan Kirstel: A podcast exploring the latest trends and innovations in the tech industry, and how businesses can leverage them for growth, diving into the world of B2B, discussing strategies, trends, and sharing insights from industry leaders!
With over three decades in telecom and IT, I've mastered the art of transforming social media into a dynamic platform for audience engagement, community building, and establishing thought leadership. My approach isn't about personal brand promotion but about delivering educational and informative content to cultivate a sustainable, long-term business presence. I am the leading content creator in areas like Enterprise AI, UCaaS, CPaaS, CCaaS, Cloud, Telecom, 5G and more!
What's Up with Tech?
Revolutionizing Wireless Communication: DeepSig on AI, Spectrum Sensing, and the Path to 6G
Interested in being a guest? Email us at admin@evankirstel.com
Discover the future of wireless communication as we sit down with Tim from DeepSig to explore the groundbreaking role of AI and machine learning in revolutionizing the telecom industry. Tim unveils how DeepSig is shifting from traditional model-driven systems to innovative data-driven approaches, enhancing modem and baseband processing, spectrum sensing, and spectrum awareness. As we edge closer to 6G, learn how AI-native functions are poised to manage the increasing complexity of wireless systems, delivering remarkable improvements in efficiency, coverage, bandwidth, and reliability. DeepSig’s flexible, data-centric methods promise to outshine conventional solutions, offering the adaptability needed for modern wireless communication challenges.
In another engaging segment, we highlight global telecommunications trends, focusing on the surging interest in Open RAN and virtualization of base station components. Learn about strategic partnerships in Taiwan and Vietnam, which underscore the critical demand for advanced spectrum performance and the integration of AI and signal classification. Discover the unmatched potential of machine learning to enhance network efficiency, mitigate interference, and boost performance for next-gen technologies like 5G Advanced and 6G. We also discuss the hurdles and opportunities of adopting machine learning in telecom, our participation in industry events such as the Mobile World Congress, and how you can join us in revolutionizing wireless technology.
More at https://linktr.ee/EvanKirstel
Hey everybody, Fascinating topic today, diving into the world of AI and ML and wireless communications technology with Tim from DeepSig. Tim, how are you?
Speaker 2:Doing well.
Speaker 1:Nice to be with you here today work you're doing as a former one-time electrical engineering student and insider in the wireless industry for 30 years Really exciting times. Before that, maybe introduce DeepSig and yourself.
Speaker 2:Yeah, absolutely so. Deepsig is about six years old at this point and we're a startup based in Arlington, virginia, and we're really focused on leveraging everything that's happening in the greater machine learning world and bringing that into some of the core wireless problems that have been problems for many years but they're getting harder and harder to deal with and to solve and so really, as we've seen deep learning transform vision and other fields, we've brought that into. You know, how do you build wireless modems and baseband processing? How do you do spectrum sensing and spectrum awareness and kind of some of those core DSP problems that you know have been a challenge for the industry for many, many years.
Speaker 1:Yeah, and enter AI and ML. What is the big idea behind DeepSig in terms of AI-native wireless technology? What are you trying to solve for exactly?
Speaker 2:So I think for many, many years, wireless has been this model-driven problem.
Speaker 2:We build a statistical analytic model for the wireless world, for how signals are transmitted and received, how we design the signals and transmit them, and I think one of the things that we've seen, you know, with deep learning and with AlexNet in 2012, is that you know we can start making this transition from really heavily model driven systems to data driven systems, and so we can let machine learning optimize directly on all of the effects you know as we see them over the air and in the real world.
Speaker 2:And so you know AI native kind of means in the wireless world. It's been used to mean you know that you're designing an algorithm or a function natively with AI, so you're learning how to process data. You know, from some end, objective, rather than you know designing an algorithm as kind of a closed form expression based on some statistical solution, analytic solution, which is how it's been done for many, many years since the beginning of wireless, essentially. So this can be applied to a lot of aspects of wireless how you process signals, how you receive signals in a base station, how you sense the signals around you in the spectrum. So, yeah, there's a lot of different ways that these kind of AI native functions can enhance wireless systems today.
Speaker 1:Fantastic, and so I've been around. I'm not sure about you, but since zero G, actually paging, I started my career in the paging industry right through to one, two, three, four, five G. So how do you see your technology shaping this evolution? You know what's next and what is your technology enable?
Speaker 2:Yeah.
Speaker 2:So I think that the spectrum is becoming really, really congested.
Speaker 2:We have a huge number of wireless systems now between our Bluetooth, our Wi-Fi, emergency communications and radars, and cellular signals and drone signals, satellite signals.
Speaker 2:We're surrounded by so many different wireless systems, so many different wireless systems.
Speaker 2:And on top of that you look at 5G, where you have massive MIMO, where you're trying to use these very complex antenna systems to multiplex many, many users into the same spectrum and dealing with orchestrating all of the different aspects of the spectrum, and you're trying to move to higher bands to lower-cost hardware. So you're trying to move to higher bands to lower cost hardware. So you have just this increasing number of constraints that you need to optimize for, and so I think one of the big things as we move on to 6G and into the future wireless systems is how do we do all of this more efficiently and how do we optimize for all of these different constraints together, you know, with real world assumptions that aren't kind of a simplified model that you get out of a textbook. I think those are some of the aspects that this can really attack and ultimately, this is improving efficiency, you know, coverage, bandwidth, reliability and allowing you to really see everything that's going on in the spectrum so you can reason about it and optimize it effectively.
Speaker 1:Fantastic. Well, speaking of textbooks, I have a few of those lying around Gathering dust on traditional signal processing methods. You know DSP from back in the day. How do you differentiate versus those traditional approaches to real-time wireless data transmission, and what is kind of unique about your architecture versus the way things have always been done?
Speaker 2:Yeah, so I think if you look at how spectrum sensing has been done in the past, there's been kind of very non-sensitive energy-based systems that do energy detection and these types of things, and then there's been very specialized systems that might be designed for one technology or a couple technologies. And really moving to a data-driven approach makes this much more flexible so you can rapidly retrain for all types of different emitters as new technologies emerge or as things like new types of interference and EMI are observed in networks. You can train for these sorts of things very quickly. So it makes it much more flexible and it can be very, you know, cheap and efficient to deploy, as we've seen kind of all of the silicon vendors going this direction. You know every, every silicon vendor is investing very heavily in, you know, efficient neural network inference and so to some degree it's riding this silicon trend as well.
Speaker 1:Yeah, absolutely Really exciting times, and there are many different approaches to spectrum sensing, different kinds of solutions and products and technologies. Your innovation is around spectrum awareness software. How do you see yourself fitting on that landscape and how are you different from all the different approaches or products out there?
Speaker 2:products out there. So today, if you look at the cellular example, monitoring for interference is really difficult. It's kind of a manual process. People look at network level KPIs. If you see throughput in a sector decreasing, you might need to send someone out with a spectrum analyzer to really sit there and do a manual process of looking for interference, hunting it down, figuring out what it is and mitigating it.
Speaker 2:All of this we can start to automate. You can have this intelligence at the edge so that your radio can figure out what's degrading it in a faster, more efficient way. And also we're seeing many, many more autonomous platforms. So we have automotive applications of wireless. We have drones, both commercial and defense-focused, that are all over the place and awareness of the spectrum is becoming more difficult because there's just such a diversity of types of devices. And also more critical, so that you can you could put this onto compact platforms and have your you know platforms be able to respond and act intelligently rather than just interfering and talking all over each other. Essentially, yeah, so we you know, yeah, so we build kind of the software core component that can be put out at the edge to provide this intelligence and then everybody can kind of build smarter systems on top of this.
Speaker 1:Fantastic. So we've seen, with the rise of 5G, you know incredible use cases for low latency communications and mission critical applications, but we're still not near you know, almost near zero latency. How do we get there? How do you achieve the really incredible real time performance and optimize for ultra low latency kind of applications?
Speaker 2:Yeah. So I think you know, as machine learning has grown across many different industries, there's been an enormous amount of R and D put into how do you build very efficient inference models, or neural network inference, and so figuring out how to decrease the size of these models into really compact, how to put them onto hardware platforms where they can have very, very low inference times, and how do you quantize and fuse these different operations so you can deploy them onto either DSPs or mobile GPUs or ASICs or FPGA devices. So we're really trying to take all of the R&D that's going on in this industry and bring it into how we deploy these models super efficiently for power and latency savings.
Speaker 1:And it can really do a lot of complexity reduction this way. Right, right, absolutely so. I've had many guests on my show, so many applications and use cases from, you know, agricultural technology to medicine and industrial Internet of Things. But what are you seeing? What are some of the industry's use cases that are benefiting from your signal classification sort of tech? Do you have any examples you can share?
Speaker 2:Yeah, definitely, you know. I think all of these industries you know, certainly medical you're seeing more and more cluttered wireless environments. Hospitals have so many wireless devices communicating over top of each other. So definitely in IoT and telecom, rapidly diagnosing and deconflicting wireless technologies is a big use case and diagnosing and mitigating any performance degradation. But there's also a major area around wireless security, so being able to understand when there's attacks going on at the physical layer in your networks or if there's unauthorized drones or other uses in the area. Certain things like data centers really want to keep out unauthorized wireless devices that could pose threats, and so that's a big area. And then spectrum really we've always just kind of assigned different bands to different technologies, but that's been really pretty inefficient. So there's really a big use case for you know how do you just do a better and smarter job of allocating spectrum so that we have more capacity where certain applications are needed and you're not wasting it on ones where you don't need them?
Speaker 1:Yeah, yeah, absolutely, and you mentioned you know so many different these variety of applications. What are the other sort of pain points that you're seeing with customers as they deploy these low latency networks? What are some of the real world challenges perhaps that you see and the practical utility you can bring?
Speaker 2:utility you can bring, yeah, I think, in the RAN space really helping to accelerate the diagnosis of, of interference or uh, both from the same tower, from different bands or from adjacent sectors or from uh. You know there's been a lot of instances of things like uh transformers or heaters, or even things like Bitcoin miners or headsets you know just emitting EMI like bitcoin miners or headsets you know just emitting emi that that can happen, you know, spuriously or hard to track times and so being able to be monitoring that kind of continuously, so you have a view into what's, what's going on in the spectrum that you can, uh, you know, act on and diagnose without a lot of resources. Uh, it is pretty um, new, a new thing for the industry that they couldn't do before. But I think you know, the biggest thing is really looking at things like 5g, where you have massive mimo, where you just have so many degrees of freedom for how you do, um, you know, beamforming, how you multiplex the spectrum to more and more users at the same time, and how you do that in the context of really complicated propagation environments with lots of buildings and reflectors and blockage.
Speaker 2:It's a hard problem to do analytically and by taking kind of this data-driven approach to it, you can really get capacity very quickly by letting it do this data-driven optimization For things. Like you know, we're doing a lot of work in Open RAN just to improve receiver sensitivity, so you get kind of extended range and coverage and capacity in the network. In the network and as we scale up to more dense multi-user systems, this can really help push a lot more out of this. And this has kind of been a notoriously hard problem in 5G. It's like how do you make the best use of all of these degrees of freedom in communication systems today? Tuning them and building these systems to operate near capacities is really a challenge.
Speaker 1:Yeah, absolutely. I'm curious on a personal note you spent a career in industry, in academia, amazing institutions like Virginia Polytech. You're an advisor to the FCC. How do you think about the industry coming together to solve these challenges? It's big challenges, but there's a lot of investment and work happening. It sounds like it's an exciting time for people like yourself sitting at the intersection of research and practical industry utility.
Speaker 2:Yeah, it definitely is. I mean, I think and I started out more on the defense side before I went into the academic space and into the telecom space and I think right now we're seeing a lot around things like CBRS, where you're sharing spectrum between federal users and commercial users, and new initiatives like the CBRS 2.0 that NTIA and FCC are leading, and so there's a bigger drive to share spectrum more efficiently here and there's also a faster path from research early research into applied research and commercial products than there's ever been. I think everything has just accelerated so much in this space that it's a really exciting time to be able to focus on these topics that are, I think, really critical to all spectrum usage right now.
Speaker 1:Yeah, and this is a global challenge, not just the US. What are some of the trends and challenges globally, and do you work with customers around the world as well as here in the US?
Speaker 2:Yeah, we definitely do. We have two major Open RAN partners with HTC in Taiwan and Viettel in Vietnam. In Vietnam there's been a big push to build Open RAN based RAN systems so that you can have virtualization of your base station components and you can maybe domestically build a number of the components, like radio units and different aspects of the stack, and so there's a massive demand now for you know, as you virtualize all of these components, how can you really make them performant?
Speaker 2:and state-of-the-art in Spectrum performance. So I think this is kind of a massive demand area that we're really focused on serving, as we work with different partners and OEMs to bring these capabilities to market.
Speaker 1:It sounds like you have a multimodal business model working with partners and customers, perhaps telcos. That must be pretty exciting, rewarding to have such an interesting bird's eye view of the industry.
Speaker 2:Yeah, absolutely. It's really awesome to be able to work with different partners across kind of the sensing and the RAN landscape to really be able to deploy some of these components into operational and deployed systems where you get the data and the ability to test these data-driven approaches in very harsh urban environments where they can really add the most value.
Speaker 1:Absolutely. What's next for AI and signal classification? What do you think is on your radar or roadmap, both from an industry standpoint, maybe academic standpoint, or even regulatory standpoint from folks like the FCC and other bodies?
Speaker 2:I think we're seeing a lot more interest in using it more broadly.
Speaker 2:I mean, it's always been fairly expensive and specialized and not super practical. We haven't had the ability to deploy sensing pervasively throughout wireless networks before, but I think just because of you know, silicon trends and these kind of algorithmic trends, it's now becoming super practical to put sensing into many different network elements, and so this is just something that's never been possible and now we can do a much better job of deconflicting interference and allocating spectrum more efficiently and improving kind of receiver and end-to-end performance. I think we're also seeing a lot of discussion in the context of of next generation, both for 5g advanced and for 6g, in terms of where aspects of this can be brought into the, the standards to really help drive performance. So things like, rather than sending channel state information from a phone back to the base station, you can learn an embedding that more efficiently and compactly feeds that back, and so this can reduce overhead and it can give you better MIMO and better beam control in a lot of these systems and I think ultimately.
Speaker 2:You couldn't want that. That is super exciting. Yeah, rather than using things like QAM and fixed framing and pilots to transmit information, you can let it learn the entire modulation and physical layer and structure to transmit information more efficiently, and so this, I think, is a really exciting concept that may find its way into 6G and for things like FR3 and these 10 gigahertz or higher centimeter bands. It can really help cope with different types of panel and propagation effects that are difficult to cope with, and to do that in a closed form way can be very computationally and suboptimally in terms of when you deploy it in the real world. So I think there's a pretty exciting opportunity here, as the industry and the standards are all kind of starting to move this direction, to try to leverage machine learning at the lower layers of the stack at the lower layers of the stack.
Speaker 1:Wow, fantastic opportunity. Congratulations on all the good work. How can people reach out to you or the team to get involved or collaborate? Are there industry meetings or events that you guys attend? Or I guess they can just go to the website, right?
Speaker 2:Yeah, absolutely. We're at deepsigai. You can reach out to us there over email or Twitter. We're also at a handful of industry events, often some of the major Open RAN events will be at Mobile World Congress again. I think this in the US, and the main one this next year, great.
Speaker 1:I'll see you there, that would be great. Yeah, we'd love to meet up there.
Speaker 2:And we're always looking for new partners and OEMs that have a need for this, because I think, as the industry starts to look at how to adopt these, it's a bit of a shock for a lot of modem and radio designers to start rapidly leveraging machine learning, which is an incredibly deep and complex field in addition to telecom. So bringing those together can be challenging and that's really our specialty is focusing on that intersect and having components that can get people going and benefiting from it very quickly.
Speaker 1:Wow, fantastic Well challenges, opportunities, disruption ahead. Thanks so much for sharing your vision and mission at DeepSig. Really exciting, Great work Onwards and upwards. Thanks, tim. Thank you, thanks for having me Take care. Thanks everyone, bye-bye.