46. 可重构智能表面准备好进入世界了吗？

欢迎来到无线未来。

Welcome, this is Wireless Future.

我是埃里克·拉尔森。

I'm Erik Larsson.

我身边是我的同事埃米尔·比尔森。

I'm here with my colleague Emil Bjarnson.

你好，埃米尔。

Hello Emil.

你好，埃里克。

Hello Erik.

我们好久没聊天了。

It's been a long time since we were talking.

是啊，埃米尔，好久不见了。

It's been a long time Emil.

所以，回到录音的感觉怎么样？

So how does it feel to be back recording?

很棒。

It's great.

所以我不得不重新 setup 我的绿幕和麦克风，这样我才能和你对话，并把这一切剪辑成一个看起来我们身处同一地点的节目，但实际上我们相距两个小时的车程。

So I had to bring up my green screen again and my microphone so I could speak with you and paste this whole thing together into an episode where it looks like we are at the same place, but we're actually in cities two hours away from each other.

哇。

Wow.

是的。

Yeah.

我的意思是，自从我们上次录制以来，发生了很多事情。

So, I mean, a lot has happened since last time we were recording, in fact.

我的意思是，当然，AI革命正在持续进行。

I mean, there is, of course, the ongoing AI revolution.

我们甚至能确定这个播客是真实的吗？确定是我们在对话，而不是某个AI读过你的书后生成的脚本？

Can we even be sure that this podcast is for real and that it's you and me talking and just not some AI script that's read your books and then created the podcast.

对。

Right.

是的。

Yeah.

人们应该留意一下我们每个人在这段视频里伸出的手指数量。

People should keep an eye on how many fingers each one of us is having during this video.

是的，一、二、三、四、五、六。

Yeah, one, two, three, four, five, six.

不，好吧。

No, all right.

那么，埃米尔，自从我们上次录制以来，无线社区发生了什么？

So what happened in the wireless community, Emil, since last time?

已经过去半年多了，

It's been more than half a year,

我觉得。

I think.

有什么新的热点话题，或者标准化方面有什么值得讨论的进展吗？

Is there anything new and hot or anything that's been happening in the standardization that'd be worthwhile to talk about?

我认为，总的来说，6G现在真的在推进，因为标准化工作已经启动，3GPP也做出了一些初步决定，也许不是关于新功能，而是关于哪些方面不应该改变。

Well, I think in general, six gs is really happening now that there is the standardization work beginning and some first decisions have been made in 3TPP, maybe not about new features, but rather about what to not really change.

例如，他们决定采用OFDM，并继续专注于5G中已有的许多其他功能。

For example, they have decided to use OFDM and continue focusing on many other features that are already there in five gs.

对。

Right.

这听起来是个有趣的方向。

So that sounds like an interesting thing.

也许我们以后可以再回过头来讨论OFDM以及这些其他波形。

Maybe it's something we should return to at some point with OFDM and all these other waveforms.

但我猜我们今天的计划是讨论另一个近年来在学术界被广泛讨论的技术，可能是过去五年、甚至十年，也就是智能反射表面。

But I guess the plan for today was to talk about another technology that's been discussed a lot, at least within academia, in the last maybe five or even years or even a decade, namely reflecting intelligent surfaces.

简称RIS。

RIS for short.

好的。

Okay.

是的。

Yes.

我想我们之前在播客里确实聊过这个话题，你和我。

I think we actually have talked, you and me on the podcast, about that.

那是在2020年播客刚起步的时候。

And it was when the podcast was in its very infancy in 2020.

自那以后发生了什么变化？

So what has happened since then?

对。

Right.

发生了很多事情。

A lot of things have happened.

我想我们上次聊这个话题的时候，它还有许多不同的名称。

I think around that time when we were speaking about it the last time, well, it still had many different names.

现在RIS已经成为大家普遍使用的缩写，但它可能意味着一些略有不同的东西。

I think now RIS has been the abbreviation that everyone is using, but it can mean a little bit different things.

我想你之前提到过‘反射’。

I think you said reflecting.

有些人让R代表‘可重构’。

Some people let the R mean reconfigurable.

以前，它也被称为智能反射面。

Before, it was also intelligent reflecting surfaces.

所以对于同一种东西，有很多不同的术语，但现在大家都称它为RIS。

So there are many different terms for the same thing, but now everyone calls it RIS.

但确实，很多事已经发生了，但核心概念仍然相同，就是拥有一个像智能镜子一样的表面，能够接收射频波，并决定如何反射它——要么将它定向成一束波，要么散射成你想要的模式。

But yeah, a lot of things have definitely happened, but it's still the same main thing of having a surface that behaves like a intelligent mirror that can sort of take radio waves that hits it and decide on how to reflect it, either as a beam in a particular direction or scatter it to a pattern that you want.

这基本上是一种能够以某种方式辅助通信系统的装置。

And this is basically something that can assist communication systems in one way or the other.

对，通过创建额外的传播路径来辅助，对吧？

Right, assist by creating additional propagation paths, right?

我的意思是，这正是它的核心理念。

I mean, that's the idea.

它通过反射信号来创建一条额外的、 presumably 强烈的从发射端到接收端的路径，从而提高信噪比，甚至在信道中增加一个额外的秩分量，以便传输更多数据。

That it would bounce the signal off or reflect the signal to create an additional, presumably strong path from the transmitter to the receiver, then boost the signal to noise ratio or even create an additional rank component in the channel so that more data can be multiplexed.

那么，到底发生了什么？

So what has happened?

因为这些想法已经存在一段时间了，过去五年里这项技术究竟取得了哪些实质进展？

Because these ideas have been around for some time now and what has really happened in the last five years with this technology?

对。

Right.

我想我是在2018年阅读资料、准备撰写一篇杂志文章时接触到这个概念的。

So I think I came across it in the 2018 when I was reading up for writing a magazine paper.

然后我认为它在2019年真正开始兴起，我们在2020年还在讨论这个话题，那时它就已经进入热潮了。

And then I think it really took up in 2019 and we were talking about this in 2020, then it was already in the hype going up.

我想现在我们可能已经超越了炒作阶段，那么人们到底取得了哪些成果？

And I think now we are probably past the hype and what has people produced?

在通信领域，已经出现了大量研究，探讨如何将RIS应用于各种系统。

Well, there have been a lot of works within the communication community, studying how you can add RIS to all kinds of systems.

当然，如果你能完美地配置它们，系统的性能会比完全没有这些装置时更好。

And of course, you do better if you're just configuring them perfectly, then things would be better than if they were not there at all in the system.

随后，人们开发出了各种信号处理算法。

Then people have developed all kinds of signal processing algorithms.

最后，还出现了大量的硬件设计。

And finally, there's been a lot of hardware design.

实际上，这种能够以受控方式反射信号的表面，早在几十年前就已在天线与传播领域被研究过。

So actually, these kind of surfaces that can reflect signals in controlled by manners have been studied for decades in the antenna and propagation community.

如今，人们已经开发出了多种原型。

And now they have been developing prototypes of different kinds.

甚至有一些公司正在销售可供购买的原型产品，以便你在自己的实验室中进行测试。

And there are even companies that are selling prototypes that you can buy if you want to test things in your own lab.

你是说，现在已经有现成的原型或产品可以购买，不仅为了在实验室里玩玩，甚至可以实际部署到现场吗？

You're saying there are like off the shelf prototypes or products that you could even buy and just for fun to play within the lab or to actually deploy in the field?

是的，正是如此。

Yes, exactly.

所以我这里正好有一个，是我从一家台湾公司T Mitek买的。

So I actually have one of those here next to me that I was buying from a Taiwanese company, T Mitek, a

哦，太棒了。

Oh, year Cool.

它能做什么？

So what is it capable of?

它长什么样？

What does it look like?

对。

Right.

这是一个方形的阵列。

So this is a square shaped array.

外框大约是28乘28厘米。

So the outer box is like 28 by 28 centimeters.

它并不大，但内部有一个18乘18厘米的阵列，包含许多可控制的小单元，用于反射信号。

So it's not that big, but at the inside there is like an array of 18 by 18 centimeters, which contains small controllable elements that reflect signals.

所以我手上的这个产品工作在28吉赫兹，属于毫米波频段，而我也可以买到工作在四到五吉赫兹的产品，那更接近WiFi频段。

So this particular product that I have is from the 28 gigahertz, so millimeter wave band, while I can also buy them for four or five gigahertz that is more like a WiFi band.

在这里，你可以拥有32乘32个可控制的单元。

And what you can do here is that you have 32 by 32 controllable elements.

也就是说，总共有1024个单元。

So that is ten twenty four elements in total.

每个单元都会反射信号，但你可以在这两个不同的相位偏移或信号延迟之间切换。

Each one of them will reflect the signal, but you can switch between two different phase shifts or delays of the signals.

所以对于每一个这样的小单元，你都可以在两个不同的相位状态之间切换，是吗？

So So individually for each one of these little atoms, can switch to phase, you said, in two different positions?

比如正负两种状态？

So like plus and minus?

是的。

Yeah.

所以大概是零度和一百八十度，或者任何其他180度的间隔。

So presumably, zero and one hundred and eighty degree or whatever other 180 degree separation.

在这里的背面，我有一个控制单元，可以连接到你的电脑，然后你可以选择想要的相位偏移模式，以控制信号的反射方式。

And on the backside here, I have a control unit that you hook up to your computer, and then you can choose what kind of phase shift pattern you want in order to control how the signal is going to be reflected.

哦，哇。

Oh, wow.

我的意思是，这个表面能重新配置的组合数量听起来简直庞大到不可思议。

I mean, it sounds like a humongous number of combinations that this surface could be reconfigured.

如果你有，比如24个单元，那么组合数就是2的24次方，几乎相当于无穷大。

And if you got, like, a 24, you said, atoms, then there will be, like, the the power of two to a 24, which is, almost like infinity.

所以，要搜索并优化所有这些组合，肯定是个极其困难的任务，我想。

So that'd be really a tough task to search through and optimize all these combinations, I can suppose.

没错。

Exactly.

但这个系统的软件简化了工作，因为它假设你希望信号从一个角度进入、从另一个角度反射出去。

But the software to this one is simplifying the life because it presumes that you want the signal to go or arrive from one angle and leave to another angle.

信号来自不同的距离。

It comes from different distances there.

然后你输入发射器的位置和接收器的位置，假设是直视场景，接着计算出你需要的相位偏移模式。

Then you type in, transmitter, this is the location to the receiver, assuming a line of sight scenario, and then you just calculate what phase shift pattern you want.

哦，所以你是说，你实际上是为特定的入射角和反射角配置这个表面，然后软件会计算出这些原子的权重。

Oh, so you're saying that you're basically configuring the surface for some specific incidents and reflection angle, and then it will compute the weights of these atoms in software.

这基本上就是你配置它的方式。

That's basically how you configure it.

是的，完全正确。

Yeah, exactly.

该

The

对，但我的意思是，如果你没有直视路径，这还能工作吗？听起来你需要直视传播才能让这个系统正常运行。

Yeah, but I mean, would that work if you don't Or it sounds like, I mean, you would need line of sight propagation for that to function.

是这样吗？

Is that so?

或者如果没有直视路径，它还能工作吗？

Or would it also work if you don't have line of sight?

没错。

Yes.

所以对于这个基础功能来说，传输首先需要满足视距条件，或者至少信号要从一个主入射角度进入，再从一个主出射角度射出。

So for this basic feature, you primarily want it to be line of sight or at least the signal arrives from one main angle and you want it to leave from one main angle.

有可能在这些特定角度范围内还存在其他一些反射物体。

It might be that there are some other reflecting objects in those specific angles.

不过对，这个功能就是专门为视距场景设计的，或是用来创建所谓的“虚拟视距”——也就是信号以视距路径到达这个表面，再以视距路径发往接收端，但这种情况里发射端和接收端可能本身无法直接互视，比如需要绕过某个障碍物。

But yeah, this is specifically for line of sight or creating what is known as a virtual line of sight where the signal sort of is coming in line of sight to the surface, going in line of sight towards the receiver, but maybe the transmitter and receiver doesn't see each other and you want to go around an obstacle, for example.

我以前听过这个术语，就是虚拟视距传输。

I heard that term before, it's a virtual line of sight.

所以你的意思是，就算原本没有直达的视距路径，只要在某处部署了智能反射面，你和反射面之间、反射面和目标之间分别存在视距路径，信号就能通过这个反射面传输，对吧？

So you're saying that means that you basically don't have line of sight, but you have a reflecting intelligent surface somewhere, and you got line of sight to the surface and line of sight from the surface, and then it kind of reflects there.

然后你就把这条带反射的传输路径等效成一条视距路径。

And you pretend that this path with reflection is like a line of sight.

对，

Yeah,

没错。

exactly.

因此，从这个角度来看，如果你想在现实世界中部署这项技术，思路是将基站固定在一个位置，同时也将可重构智能表面（RIS）部署在固定位置。

So from that viewpoint, if you want to deploy this in the real world, the idea is that you have a base station at a fixed location and you deploy this RIS also at a fixed location.

在部署阶段，你可以确保它们彼此可见，但接收端可能位于多个固定位置，你需要在它们之间切换，或者接收端可能是移动的，因此你需要调整信号反射的方向，以便覆盖到用户。

And there you can already make sure that they see each other in your deployment phase, but then the receivers might be at multiple fixed locations that you want to switch between or they can move around and therefore you want to change where you are reflecting the signal so you are reaching the users.

对，当然。

Right, sure.

所以你桌上这个小设备，当移动设备移动时，它能否自动重新配置？还是说每次都需要重新修改软件？

So this little product that you got on your desk, would that be capable then of reconfiguring itself when the mobile maybe moves around, for example, or would you have to like redo the software every time?

这到底是怎么运作的？

How does that work?

对。

Right.

到目前为止，在我自己的实验中，我主要用这种方式，通过以太网线连接并用你的

So so far in my own experiments, I only use this more at the slow pace where you connect it with an ethernet cable and you controlling it with your

自己用键盘控制。

keyboard yourself.

但你也可以连接一根电缆，配合某种实时算法使用。

But there is a cable that you can connect as well that you can use together with some kind of real time algorithm.

我见过其他实验室的人这么做。

And I've seen other people in labs doing that.

然后你可以大致估算用户的位置，并将数据输入进去。

And then you can basically estimate where a user is and feed it in there.

但当然，这种估算本身也是一个挑战。

But of course that estimation is also a challenge on its own.

当然。

Sure.

是的。

Yeah.

我很想知道更多关于它的硬件事实。

I'd be curious to know more about its hardware, fact.

但这些RIS设备将用于什么用途呢？

But what are these RIS gadgets going to be used for?

我的意思是，经过这么多年的研究，现在它们最主要的使用场景或杀手级应用是什么？

I mean, what are the prime use cases or killer applications for them that have emerged now after all these years of research?

对。

Right.

所以我们最近写了一篇论文，题为《重构智能表面在60GHz中高频段：收益还是痛点？》

So we wrote a paper recently called Reconfigurable Intelligent Surfaces in Upper Midband 60 Networks Gain or Pain?

收益还是痛点，还是痛点还是收益？

Gain or pain or pain or gains?

是的。

Yes.

是吗？

Is it?

所以中高频段是指毫米波吗？还是中高频段指的是什么

So the upper mid band is like millimeter wave or what is the upper

中频段？

mid band?

这实际上是我们今天在3.5吉赫兹频段使用的频段之间的范围。

That's actually the range between what we are using today in the 3.5 gigahertz band.

基本上从7到28或24，这就是所谓的上中频段。

So basically from seven up to 28 or 24, that is what's called upper mid band.

而人们特别瞄准这个频段来构建新的系统。

And that is what people are targeting particularly to build new systems there.

所以如果你要构建新系统，你也可以从一开始就规划部署你的RIS。

So if you build new system, you also could have the opportunity to plan from the beginning to deploy your RIS.

对，从7到28。

Right, so from seven to 28.

我的意思是，这听起来主要是一种用于更高频率的技术，但据我了解，毫米波接入在5G中并没有那么成功。

I mean, sounds like it is primarily a technology for higher frequencies, but also to my understanding millimeter wave access for example was never so successful in five gs.

所以它是否也能用于较低频率，或者只是会变得太大太重？

So maybe could it be used also for lower frequencies or would it just become too big and heavy?

那具体是怎么回事？

How is that?

我的意思是，为什么你们在论文中研究的是这个频段？

I mean, how is it that this is the band that you studied in your So paper

我们在这篇论文中研究这个频段的主要原因是，现在大家正有兴趣在这一频段部署网络。

the main reason that we studied in this paper was that this is just what people are interested in deploying your network in that band.

但我认为，原则上你可以在任何频段部署这些智能表面，但有大量迹象表明，当你提高频率时，RIS相比其他提升覆盖的方式能带来更大的优势，特别是因为我们知道，频率越高，覆盖就越呈现二元性——要么有覆盖，要么被大型物体阻挡。

But I think in principle, you can deploy these surfaces in any frequency bands, but there is a lot of indications that you are benefiting the most from RIS as compared to other ways of providing better coverage when you go up in frequency, particularly because we know that the higher up in frequency you are, the more the coverage becomes kind of binary that either you are in coverage or you are blocked by big objects.

而RIS可以产生巨大影响。

And then an RIS can make a huge difference.

对。

Right.

我能理解，因为随着频率升高

I can see that because as you go

频率越高，传播特性就越少散射，更像是光束，要么能穿透，要么被阻挡，或者在某处被反射。

up in frequency, propagation becomes more like less scattering maybe and more like optical rays that either make it through or don't or get reflected somewhere.

那么在你们的论文中，研究和模拟了哪些使用场景呢？

So what are the use cases that you in your paper have studied and simulated and So discussed,

我们研究了四个我们认为对未来有重要意义的特定使用场景。

we studied four particular use cases that we think are relevant for the future.

而且，是的，这些场景可能不仅适用于单一的载波频率。

And yeah, it could be relevant for more than just one carrier frequency.

但第一个是固定无线接入，这是一个在4G时代之前我们并没有真正关注的场景，但在5G中却成为了一个重要的新用例。

But the first one is fixed wireless access, which is a use case that we didn't really have before four gs, but it turned out to be one of the important new use cases in five gs.

即使人们在讨论5G时提到了许多其他方面，但在一些光纤入户率不高的市场，固定无线接入变得非常重要。

Even if people were talking about many other things in five gs, this is actually one that on some markets where we don't have so much fiber penetrations to people's homes, fixed wireless access became an important thing.

固定无线接入。

Fixed wireless access.

哇。

Wow.

所以是在7到28吉赫兹这个频段？

So in this band from seven to 28 gigahertz?

是的，原则上是这样，但你可以提供任意你想要的带宽。

Yeah, in principle, but it could be any bandwidth you would like to provide it.

所以这是一种场景，你不需要挖光纤到用户家中，而是希望通过无线方式提供类似的性能。

So this is a scenario where instead of digging a fiber cable to someone's home, you would like to deliver similar performance wirelessly.

然后你可以在用户的家中安装一个天线。

And then you can put out an antenna at the home of a user.

接着，从基站出发，你希望覆盖尽可能多的住宅，以达到一个关键数量，这样对基础设施的投入就会更值得。

And then from a base station site, you would like to provide coverage to as many homes as possible so that you can reach a critical number so this investment in more infrastructure will be better.

然后，如果你部署一些可重构智能表面（RIS），就可以扩展基站的覆盖范围，覆盖另一批不同的住宅。

And then the idea would be that if you put out some RIS, well, then you can extend your coverage at this base station and cover another bunch of different homes.

我明白了。

I see.

但RIS会安装在哪里？

But where would the RIS be installed?

它会靠近用户天线所在的住宅吗？

Would it be close to the home where you got like the customer's antenna?

还是它更靠近基站？

Or would it be closer to the base station?

还是它位于两者之间的某个位置？

Or would it be somewhere in the middle between the two?

对。

Right.

所以从数学上讲，如果想获得最佳的信噪比，并且可以自由选择位置，那么它应该靠近发射端或靠近接收端。

So mathematically, if one would like to get the best signal to noise ratio, and if you have the liberty of picking anywhere, well then it should either be close to the transmitter or close to the receiver.

我可以想象，为了服务众多不同的客户，你可能会选择一个山谷之类的地方，从基站能看见山谷的顶部，然后在那里部署一个可重构智能表面，让它将信号反射下去，快速切换指向不同的房屋，从而在这些家庭之间分配你的容量。

I could imagine that in order to get many different customers here, you would like to find a place that is maybe a valley or something like that, that you can see the top of this valley from your base station, you put an RIS there and then you let the RIS sort of reflect the signal down towards different houses switch between them quickly so that you can provide and divide your capacity between these different homes.

所以当你说到‘切换’时，你的意思其实是随着时间切换，让这些用户实现时分复用，而在某一特定时刻被选中的用户会获得信噪比和数据速率的提升。

So when you say switch between them, you really mean like switching over time so that these users are time multiplexed and whoever gets favored by the reset the very specific moment will get a boost in SNR and data rate.

是的，完全正确。

Yes, exactly.

事实上，我以前从未考虑过这种应用场景，但我想在固定接入中，根据名称本身就能大致推断出来，客户天线和基站都是静止的。

Fact, I never thought about that use case before, but guess in fixed access, this is implied by the name more or less, the customer antenna and the base station will be stationary.

因此，你只会有一些来自RBAC的微弱反射，这些反射可能是移动的，但在高频载波下，它们对信道的贡献可能很小，会长时间保持静态，因此配置RIS会相对容易。

So you'd only have a few small reflections from RBAC to run that might be moving, but at high carrier frequencies those might not contribute a lot to the channel, will be static over a long time, and then it should be comparatively easy to configure the RIS.

我的意思是，你可以大致告诉它你希望的入射角和反射角，然后就是时间切换模式，我想。

I mean, you could tell it more or less the angle of incidence and the angle of reflection that you desire and then just the time switching pattern, I suppose.

这样是可行的。

That would work.

是的，没错。

Yeah, exactly.

因此，算法也需要根据你当前所处的特定传播场景进行调整。

So that also comes into them that the algorithms are important to be tuned to the particular propagation scenario that you're having in this case.

这条直视路径是变化缓慢的，因此我们可以针对它进行设计。

This line of sight, it is something that varies slowly, so we can design for that.

而且，像这样的表面本质上只能从一个方向接收信号，并且只能向一个方向反射，或者至少在同一时间只能在整个频段内保持一种反射模式。

And then it is an important thing that a surface like this can basically only take a signal from direction and reflect it in one direction, or at least it can only have one reflection pattern over its entire frequency range at the same time.

因此，你需要对用户进行时间复用，而不是在频率上分割它们，或者同时发送多个波束给多个用户。

And that is why you need to time multiplex users rather than split them in frequency or send multiple beams to use at the same time.

通过RIS，你只能提供一个波束，但可能存在其他场景，其中已经有一些路径指向用户。

Well, through an RIS, you can only provide like one beam, but there can be other scenarios where you already have some paths directed to users.

然后引入ARIA，可以增加一个额外的路径，帮助你区分不同用户。

And then when throw in the ARIA, get an additional one that can help you distinguish between the users.

对。

Right.

那么，你还能想到其他哪些应用场景呢？

So what other use cases do you foresee for

这些？

these?

对。

Right.

在仿真中，当我们涉及固定无线接入时，最大的增益出现在我们之前提到的这种虚拟直视路径场景中。

So what we really saw in the simulation when it come to this fixed wireless access is that the biggest gains comes in scenarios where you create this kind of virtual line of sight path that we talked about earlier.

因此，你需要与表面保持直视，并且希望RIS到各个家庭之间也保持直视。

So you need to have line of sight to the surface and you want to have line of sight from the RIS to the different homes.

还可以想象其他一些场景，在这些场景中，你希望在小区的特定区域提升容量。

And one can imagine other scenarios where this will happen, some situation where you would like to enhance capacity within the cell in particular regions.

例如，你可能很好地覆盖了某个区域的街道，但你希望信号也能传到其他街道，或者在某些地方有大型遮挡物，你希望将信号反射到这些被遮挡的区域。

So, for example, you could have a situation where you are covering nicely the streets in an area, but you then would like the signals to go into some other streets or yeah, you have some big blocking objects somewhere and you want to reflect into those areas that those objects are blocking.

但现在，与固定位置不同，用户会在这些区域中移动，因此你希望更动态地调整RIS配置，以便跟踪这些用户。

But now instead of having fixed locations, you have users that moves around in these areas and therefore you want to more dynamically change the RIS configuration so we can follow those users.

你现在谈的是移动接入。

Now you're talking about mobile access.

而此时，客户可能是手机之类的设备，在小区内快速移动。

And now we might have like the customer here might be a cell phone or something that's moving around quite fast in the cell.

是的，毫无疑问。

Yes, definitely.

当然，如何实时重新配置你之前提到的所有权重，将是一个挑战。

Then of course there will be a challenge on how to reconfigure in real time all these weights that you talked about earlier.

是的，正是如此。

Yes, exactly.

这 definitely 会成为一个问题。

So this is definitely going to be an issue.

关于这一点，有趣的是我们再次发现传播场景的选择会产生巨大差异。

The interesting thing when it comes to this is that we once again discovered that the choice of propagation scenario makes a huge difference.

首先，ARIES在具有直接视距的情况下效果最明显。

First of all, the ARIES is making the biggest difference when you have line of sight to and from it.

因此，你主要帮助的是那些能够看到ARIES的用户，即使他们正在移动。

So you are helping mainly users that can see the ARIES as well, even if they are moving around.

这在实际进行配置时为你提供了很大帮助。

And that is helping you a lot when it comes to actually doing configuration.

因为如果你设计一个智能的估算算法，就可以将场景缩小到：我知道我的基站物理位置。

Because if you design a smart estimation algorithm, you can then narrow down the scenario to, okay, I know physically where my base station is.

我知道RIS的位置。

I know what the RIS is.

我想估算的是RIS到用户的角度，以及可能的距离。

What I like to estimate is sort of the angle and maybe the distance from the RIS to the user.

这只需要两个自由参数，而相比之下，我旁边这个ARIA有1024个单元。

And that is just two free parameters as compared to, yeah, you have a ten twenty four elements inside the ARIA I have next to me.

因此，你可以将需要估计的参数从上千个减少到仅两个，这在原则上要容易得多。

So, you can reduce everything from a thousand to two free parameters to estimate, which is much easier in principle.

我明白了。

I see.

我的意思是，听起来像是

I mean, sounds a

有点像任何MIMO系统中的信道估计——如果你知道你处于直视路径或远场直视路径，那么只需估计角度或可能的俯仰角就足够了。

bit like channel estimation in any MIMO system where, of you know that you're in line of sight or you're in far field line of sight then it would be enough to estimate the angle or possibly elevation and estimate.

但无论如何，只需估计少数几个参数，而不是成千上万个系数。

But in any case just a few parameters as opposed to estimating thousands of coefficients.

那么增益有多大呢？

So how large are the gains then?

我的意思是，如果我们把ARIES部署到一个有移动用户的蜂窝系统中，我们预期会看到什么样的效果？

I mean, what would we expect to see if we were to like deploy ARIES in a cellular system with mobile users?

我的意思是，假设我们有较大的带宽，因为我们通常确实如此，对吧？

I mean, let's say that we've got a large bandwidth because we typically do, right?

可能是20兆赫，甚至可能达到100兆赫左右。

It could be 20 megahertz, could be even, I suppose, up to like 100 megahertz or so.

这些用户难道不是已经获得了他们所需的所有容量吗？

Doesn't this user don't these users already get all the capacity they need?

或者RIS还能额外提供多少容量？

Or how much could the RIS provide additionally?

对。

Right.

这在很大程度上取决于具体场景，但在我们的仿真中，假设了100兆赫的频谱，用户的性能范围在每秒每赫兹4到10比特之间。

So this becomes very scenario dependent, but in our simulations here, assumed 100 megahertz of spectrum and then we had a range of performance from four to 10 bits per second per hertz for the users.

因此，在这种情况下，每个用户可以额外获得1到2比特每秒每赫兹的容量。

So we could add one to two additional bits per second per hertz per user in this situation.

所以，相对而言，你可能会获得大约50%的额外增益，或者说每个用户能多获得几百兆比特每秒的速率。

So, yeah, relatively you could get some 50% extra or you can just say that every user can get a few 100 megabit per second extra.

不，用户可能并不需要所有这些额外的容量，但当系统需要在大量用户之间频繁切换时，系统本身可能需要它。

No, probably users don't need all of that extra capacity, but the system itself might need it when you have a lot of users that you would like to switch between.

因此，你可以更侧重于提升系统的整体容量，而不是单纯地提高每个人的性能。

And so you can more enhance the capacity of your system rather than maybe boost the performance for everyone.

但在小区边缘的一些场景中，用户确实会明显受益，性能会从几乎为零提升到相当不错的水平。

But there will be some use at the edge of the cell that definitely benefit and it goes from really no performance to decent performance.

我能想象，毕竟在无线通信中，覆盖小区边缘总是很困难的，对吧？

I can imagine, I mean, it's always covering the cell edge that's difficult in wireless comms, right?

而且，小区边缘也不一定非得理解为离基站很远。

Also cell edge is not to be necessarily taken literally that you're far from the base station.

你可能处于一种不利的环境，比如被障碍物遮挡、身处地下室深处，或者即使离基站并不远，但由于某种原因，你与基站的信道质量很差，反而与相邻或其他小区的干扰源信道更强。

It could be that you are really in an unfavorable situation where you're blocked by some objects, you're deep in the basement or somewhere, or even that you, for some reason, have you might not be very far away again from the base station where you could have a very poor channel to the base station and a much stronger channel even to an interfering to to the next or other cell.

我认为，正是在这些情况下，RIS能够显著提升信噪比，从而发挥潜在作用。

And I suppose it's in those situations where you could really boost the signal to noise ratio and where the RIS could potentially help.

这将非常有价值，因为这正是传统技术难以有效覆盖的场景。

And that would be valuable because that's really the scenario that's difficult to cover with standard technology.

那么，还有其他情况吗？比如，你是否能期待更大的提升？

So are there other situations, I mean, where you could expect even larger gains?

也许并不是更大的提升，但你可以想象，在小区内也有一些特定区域。

Maybe not larger gains specifically, but you could imagine that there are particular regions in the cells as well.

比如，有一个公园，那里有很多用户，或者一条邻近的街道，你希望在那里提供更好的覆盖。

Like there's a park where you have a lot of users or a neighboring street or something like that where you like to provide better coverage.

但我们真正注意到的是，在大量研究RIS的学术工作中，他们都说：‘我唯一能连接到用户的方式就是通过RIS。’

But one thing that we really saw was that in a lot of the academic work that are studying RIS, they say, Oh, the only way of reaching my user is through the RIS.

发射机和接收机之间没有任何直接路径。

And there is no path directly between the transmitter and the receiver.

完全是这样，

It's totally Oh,

等等，等等，我不太明白。

wait, wait, there's no, no, I'm not getting it.

发射机和接收机之间不是总有一条直接路径吗？

So there's always a path between the directly between the transmitter and the receiver, isn't there?

我的意思是，它可能很弱，但总是会存在的。

I mean, it might be weak, but it's always going to be there.

不是吗？

No?

是的，正是如此。

Yes, exactly.

这正是我们非常清楚地看到的要点。

That is the point that we were seeing very clearly.

人们经常说它非常弱，因此可以忽略它。

People often saying that it's very weak, so therefore we can neglect it.

但我们发现，在中高频段，如果忽略这条路径，整个仿真结果都会出错。

But we see that it actually in the upper mid band plays So a huge if one is neglecting that path, everything will be wrong in the simulation.

所以我们注意到了这一点。

So that was something we noticed.

我可以

I can

听到这一点。

hear that.

但它也点明了这一点。

But it then also puts the finger on that.

在现实中，我们预期发射器和接收器之间会存在什么样的静态路径？

What kind of static path between transmitter and receiver would we expect in reality?

嗯，在这种场景下，信号需要穿过建筑物，或者在到达接收器之前经过多次反射。

Well, maybe in this scenario where the signal needs to go through the building or bounce on a lot of objects before it reaches the receiver.

如果我们再提高频率，比如使用毫米波，就像我这里提到的RIS，那么相对而言，这些路径通常会变得弱得多。

And if we then go up in frequency, two millimeter waves, for example, as the RIS I have here, well then relatively speaking, those paths typically becomes much weaker.

因此，我们也可以想象，RIS在更高、更混响的频率下会更具吸引力，不是因为反射本身有显著不同，而是因为即使没有RIS，那些静态路径也会变得微弱得多。

So we can also imagine that RIS becomes more attractive to higher reverberating frequency, not because the reflection as such becomes much different, but because these static paths that will be there, even if the RIS is not there, day becomes much smaller.

因为静态路径变弱了。

Because the static paths become smaller.

我明白了。

I see.

不，很有趣。

No, interesting.

当然。

Sure.

因此，有人可能会认为，RIS在毫米波甚至更高频率的场景中最为有用，因为那时每个基站的覆盖范围都非常有限。

So what one could potentially argue that RIS is mostly useful in situation where you are at millimeter waves or even higher frequencies, because then each base station has a very limited coverage.

而有些人有时会争论说，如果覆盖不好，就多建一个基站。

And if you want argument people are having sometimes that if you have bad coverage, just put up another base station.

但如果你再建一个基站，那么你还是会面临另一个覆盖范围有限的问题，而相比之下，在较低频率下，一个基站就能实现360度全覆盖。

But if you put up that one, well, then you would have yet another place where you have limited coverage as well as compared to putting up at lower frequency where you can get three sixty degree coverage from one base station.

因此，在高频段部署几个RIS，其效果会远大于在低频段部署。

So a few RIS might make a much bigger difference at higher frequencies than it would do at lower frequencies.

那么在实际中部署这些RIS有多容易呢？

So how easy would it be to put this up in practice?

我的意思是，应该不需要特别复杂的安装吧？

I mean, wouldn't need to be installed, I suppose.

它们需要电力。

They would need power.

它们还需要回传链路。

They would need any back haul.

不用。

No.

但它们需要电源，而且 presumably 一些控制信号会通过无线方式传输。

But they would need a power supply and presumably some sort of control signaling would be through the air.

所以它们只需要电源和一个可以安装在墙上的地方即可。

So all they need is a power supply and somewhere to sit on a wall or somewhere.

是的。

Yes.

我的意思是，与额外部署一个接入点相比，即使是在户外，它们的安装似乎要简单得多，成本也低得多。

So I mean, it sounds like they would be a lot easier to and cheaper to install than putting an extra access point there, even outdoors.

这么说准确吗？

Is that accurate to say?

是的，这一直是我们的愿景，我认为这确实是正确的。

Yes, this has been the vision all the time and I think it seems to be true.

但确实，其中一个问题是电源供应。

But yeah, one is the power supply.

我认为这个原型的功耗大约是二十到三十瓦。

I think this prototype here consumes on twenty, thirty W.

但如果你真的把它设计得更精简，功耗可能会远低于这个数值。

But if you really build it to be more lean, then it could be substantially less than that.

但确实，你需要某种电源供应，得考虑如何将其部署在墙上之类的地方，以及如何旋转它，因为和其他任何射频设备一样，你希望它体积大一些并能旋转，以便其覆盖角度不会像杜恩那样过高。

But yeah, some power supply you would need for is you need to figure out how to deploy it maybe on a wall or something like this and how to rotate it because just as any other RF hardware, you want it to look big and rotate it so that the angle that it reaches it are not too high as compared to the Duhrn.

是的，我的意思是，20瓦仍然相当可观，但如果我们能像你说的那样设计得更节能，也许功耗可以降到几瓦左右，甚至可以用太阳能供电。

Yeah, I mean 20 watts is still substantial, but if we could build it, as you said, leaner, maybe it could consume like a handful of watts or so, it could even be solar powered.

这很可能是一个合理的论点。

Probably a case could be made for that.

但与此同时，我们很难与网络控制中继器竞争，后者结构简单得多， presumably 更便宜，也更节能。

But at the same time, it's a little difficult to see how we could be competitive to network control repeater, which is a lot simpler to build and presumably cheaper and also more power efficient.

但那是另一个话题了。

But that's another story.

好的。

Okay.

我认为这里消耗最多的其实是控制单元，因为我早年也参与过一篇展示RIS户外应用的论文。

And I think it's the controlled unit here that consumes most of it because I was also part in one of the papers that demonstrated outdoor RIS usage at the early years of this technology.

在那里，我认为RIS本身消耗不到一瓦，但控制单元消耗了大部分功率。

There, I think the RIS as such was consuming less than a watt, but the control unit consumed most of it.

那只是因为当时还没有针对这一点进行优化。

That was just because it wasn't optimized for it.

哦，真的吗？

Oh, really?

因为当时没有优化。

Because it wasn't optimized.

我能理解。

I can imagine.

是的

Yeah.

当然

Sure.

好的

Alright.

我们之前讨论过配置所有这些单元的挑战。

So we talked earlier about the challenge to configure all these atoms.

在你的原型中，你用了24个。

And you had in your prototype, you had a 24 of them.

每个单元都需要配置一个相位偏移，而且是离散的相位偏移。

So each one needs to be configured with a with a phase shift and with a discrete phase shift.

我想你说过它是二进制的，基本上就是正负一。

I think you said it was like binary, it's basically like plus minus one.

但在我看来，让RIS实际运行并高效工作的真正挑战在于：如何基于实时信道状态信息估算出这些权重应该如何设置，然后立即把相关信息传达给RIS。

But to me, this is the real challenge in making RIS operational and efficient, how to actually estimate channel state information in real time based upon this channel state information compute how these weights should be set and then also communicate that information instantly to the RIS.

这不会是一个巨大的挑战吗？

Won't this be a big challenge?

特别是如果你在小区边缘，信号会很弱，控制信道 presumably 也会很弱。

Particularly if you're at the cell edge, you're going have a weak signal and the control channel will also be weak presumably.

至少如果这个控制信道是在同一频段内发送的话。

At least if that control channel is sent within the same band.

我不确定。

I'm not sure.

也许它可以通过带外方式或通过电缆等其他方式发送。

Maybe it could be sent out of band or through some other means through a cable maybe.

我的理解是，至少某人的控制信令是在频段内进行的。

My understanding has been that somebody's control signaling at least would happen within the band.

要实现这一点，难道不会是一个巨大的挑战吗？

Won't that be a huge challenge to pull off?

是的。

Yeah.

我认为这在现实世界中仍有待充分验证。

And I think it is something that remains to be fully demonstrated in the real world.

理论上，如果信道完全没有结构，你就需要估计1024个参数，必须至少发送1024次信号才能观测到所有不同参数并进行估计。

In theory, the problem would be that if there is no structure at all in the channels, you have ten twenty four, in this case, entries to estimate and you would have to sort of send signals at least ten twenty four times in order order to observe all the different parameters and estimate them.

这将花费大量时间，而且信道可能已经发生了变化。

And that will take a lot of time and maybe the channel will have changed.

当我们研究MIMO通信中的类似问题时，你希望基站能够覆盖世界任何地方。

And when we worked on similar kind of problems in MIMO communications, then you want the base station to be able to cover any part of the world.

因此，你必须以某种方式解决这个最具挑战性的问题。

So therefore you need to solve that most challenging problem somehow.

但当谈到RIS时，如果我们把使用场景缩小到这种虚拟视距情况，仅需估计角度。

But when it comes to the RIS, if we narrow down the use cases to this kind of virtual line of sight situations where you only want to estimate angles.

那么，你至少可以简化你的算法。

Well, then you can at least simplify your algorithms.

然后，你主要就面临你提到的信噪比问题。

And then you mainly have this issue with the SNR that you mentioned.

例如，许多算法建议让用户发送几次已知信号。

For example, what a lot of the algorithms are suggesting is that you let the user send the known signal a few times.

RIS会在一些预设配置之间切换，以不同方式估计或观测信道。

The RIS is changing its configuration between some predefined ones in order to estimate or observe the channel different ways.

然后你有

Then you have

一个算法试图预测角度，计算你应该采用哪种配置，然后你可以在基站进行这种计算，并让基站通知RIS。

an algorithm that tries to predict and oh what was the angles, calculating what configuration you should have, and then you can do that calculation at the base stations and you can let the base station tell the RIS.

不，这说得通。

No, but that makes sense.

我的意思是，这本质上是在利用结构，甚至利用传播环境，这样你就不必尝试所有可能的组合，而只需选择少数几种。

I mean, basically it would be exploiting structure or even in the propagation environment so that you don't have to try all possible combinations, but just a select few.

然后基于这些，推断信道的状况。

And based upon that, perform some inference on what the channel looks like.

再根据这种推断，计算应该如何设置。

And then based on that inference, compute how the it should be set.

但这么说起来，它似乎是一种扩展性不好的技术。

But it still sounds like a, in a way, a technology that scales poorly.

我的意思是，如果你从100个增加到1000个、10000个，甚至百万个RIS中的原子，恐怕需要相应地增加大量信令。

I mean, if you were to go from like a 100 to a thousand to 10,000 or or, you know, a million or something atoms in the RIS, would need correspondingly more signaling presumably.

因为原子越多，你对RIS效用的期望也就越高，特别是对它所能带来的信噪比提升和功率增益的期望。

Because it's also that the more atoms you put there, the higher would be your expectations on the utility of the RIS, and specifically on the increase in SNR and power boost that it would give.

而信噪比越高，你越可能发现更多的散射点，导致你的模型开始崩溃。

And the higher you go in SNR, the more scattering points you're likely to discover so that your model starts breaking apart.

我的意思是，在低信噪比下，你或许可以说只有直射路径。

I mean, you're at low SNR, could say there's only line of sight.

当你提高场景的信噪比时，你会发现实际上还多了一条反射路径。

You go up within the scenario, you discover there's actually a reflected path on top.

再进一步提高，你会发现又多了大约十个反射或散射点，这些也必须考虑进去。

You go up a little further, you discover that, oh, now there are like another 10 reflections or scattering points that we also have to consider.

所以，对我来说，它是否能很好地扩展并不明显。

So it isn't obvious to me that it scales very well.