Good night. Maybe the concepts are different. But I have seen something similar, in Lampsite (small cells from provider X), where they could define several pRRU (pico Remote Radio Unit) related to the same cell. The connection is with PoE.
Similar concepts have been discussed for 20 years. There were a category of collaborative methods between access points (remote radio units) in the 4G era called “coordinated multipoint”. What we talk about in the context of cell-free massive MIMO is the method called “joint transmission/reception”. Huawei has a product that they call LampSite which seems to provide similar features for indoor use. I am not familiar with what kind of algorithms they are using and whether it matches with cell-free massive MIMO, or is more simplified.
Hi Professor, thank you for the informative video. I have a question regarding the Uplink Spectral Efficiency (SE): Why does the scenario where all APs serve all UEs yield better results than the DCC (Dynamic Cooperation Clustering) version? I would think that having all APs serve all UEs might cause more interference. It seems that the DCC version could potentially achieve better SE by reducing interference through selective AP-UE connections. Could you please clarify this? Thank you!
If the uplink processing is not designed to deal with interference, then some APs can cause more harm than good, in the way you described. You can find numerical examples of this in the paper “Cell-Free Massive MIMO: User-Centric Approach”. However, that cannot happen if the uplink processing is optimally designed. When all APs serve all users, you have more options available so you can choose to disregard the received signal if it contains more interference than useful signals. Since you have more information available when you consider the received signals from all APs instead of only a subset of them, you can implement the same receiver processing and more, which is why the optimal solution cannot be worse.
Good explanation! A small question, at around 8:15 , you mentioned that in TDD mode, the user send pilots and base station estimate the channel. Then in the downlink, should the base station send pilots and user estimate the channel? ❤ Thanks!
This is a good question that I tried to answer in a blog post a few years ago: ma-mimo.ellintech.se/2018/11/02/when-are-downlink-pilots-needed/ From a theoretical standpoint, it is sufficient to send pilots in one direction and then use it to make the effective channel predictable in the other direction. However, in practice, we often send pilots in both directions to deal with hardware impairments. Insufficient calibration between the hardware components used in uplink and downlink will lead to slightly different channels. This is particularly the case with phase drifts that are added with different signs in the uplink and downlink.
@@WirelessFuture Thanks for your explanation! In this blog, it is assumed that uses are equipped with single antenna and downlink pilots are not needed theoretically. For the case where multiple antennas are equipped, should we send downlink pilots? I think users must know h to combine signals ( MRC et al.) and hence downlink pilots are needed. Is that true? Additionally, if downlink pilots are needed, the number of downlink pilots is proportional to the number of antennas at the base station M, which may incur high signaling cost in massive MIMO, right?
You are right that downlink pilots will be needed whenever the user device has multiple antennas. In that case, one should transmit one pilot per downlink beam, and beamform/precode it as one will do what the downlink signal. The number of downlink pilot is then equal to the number of beams, which is typically much lower than the number of base station antennas. So the signaling code is not overly high - typically the same as the uplink pilot signaling.
Great Illustration! Regarding the pilot channels, I have a question: the Master AP listens and tries to find out which pilots are received with the least interference, so in other words you are pointing out to the pilot channel which is not chosen " that much" by the "slaves" AP's to serve certain UE's in which they form a Master AP with respect to them ?
@@WirelessFuture But if may I ask, do you think that the Master AP listening to the pilot signals transmitted from other AP's on a certain resource set would guarantee that the next transmission instance of this resource set would lead to similar interference level at the time where the chosen pilot shall be transmitted? Kindly correct me if I'm mistaken, the resource sets can be changed from one instance to another, when it comes to the number of RS they hold, which in turn depends on the number of UE's which the AP is trying to serve, which is also varying(The number of UE's under one Master AP) ? Isn't it better to centralise the pilot status of the AP in each cluster at the centralised controller which can eliminate this kind of ambiguity ?
Do you only assume one CPU? What happens then if there is more than one CPU? In one of your publications, you mentioned that the real case ( in practice) is a system with multiple CPUs.
We are normally considering that there is one CPU with access to all information that the APs are sharing. However, it doesn’t have to be one physical entity. The CPU tasks could be distributed between multiple servers. We mention this briefly in Chapter 1 of the book arxiv.org/pdf/2108.02541.pdf One can also optimize the allocation of the CPU tasks over physical processing units, as in this paper: arxiv.org/pdf/2301.06166.pdf
Hello professor. I have a question that is bothering me lately. When I search the literature I find that 95% of the works use only spectral efficiency/rate as a performance metric and nobody consider the BER as a performance metric. No doubt the capacity say a lot about the performance but ultimately the bits need to be correctly decoded at the receiver. In that case why is it that nobody is giving any attention to improve BER? I'm working on CF-mMIMO-OTFS and even the first paper on CF-mMIMO-OTFS has mentioned that the BER is an important metric but they left it as a future work. I tried simulating CF-mMIMO-OTFS and I found that the BER is good as long as the number of users are restricted to 2. But as soon as I increase the number of users to more than 2, the interference power increases significantly because of which the BER degrades significantly as well. Can you please give your insights on how to improve this? Perhaps NOMA can help? I'm not sure. Thanks in advance
This is a great question that I have received many times. Maximizing capacity and minimizing BER often go hand-in-hand from an overall design perspective (i.e., how to point beams). However, in practice you often pick a modulation/coding scheme that gives an acceptably small BER, so it is not of a constant than a variable. I elaborate more on this in a new video (please feel free to ask follow up questions related to it): Bit Error Ratio or Channel Capacity: Which metric should I use? [Frequently Asked Questions] ua-cam.com/video/sK537pnqjWM/v-deo.html
@@WirelessFuture Dear professor, thank you for responding it with a video. I agree with your point but then again it is true provided the detection is done correctly. What I do not fairly agree with some of the works is especially when they show capacity plots for SIC-based MMSE detection in CF-mMIMO when in reality SIC never works in cases with more than 2 users. When we can't even do detection this way and can't get a good BER, isn't it redundant to present the capacity plots in this case? I feel both capacity as well as BER together should be considered as a performance metric.
There are no decoding error when considering the capacity - it is achieved with zero BER (and even zero error in the absolute sense) as the block length goes to infinity. However, in practice, there will certainly be decoding errors and SIC makes the situation worse through error propagation. It is not the BER but the block error probability that matters in practice: What is the probability that a coded data block is decoded correctly? If it is decoded correctly, we have BER=0. If it is decoded in error, it doesn’t matter what the BER is because you anyway need to retransmit the data.
I imagine for this to work the each site has to be synchronized to an extremely high degree. I would presume GNSS would be the standard go to but several nations have indicated that they are more than willing to turn the systems off if they feel the need to do so additionally these systems are susceptible to jamming and spoofing. They also need high bandwidth low latency site to site traffic if they are to do tricks such as computational interference subtraction, I think this would be orders of magnitude more than the actual data you are interested in, you probably need the raw IQ data. How long is a 5g frame? Will each site need to be meshed together with fiber or will microwave backhauls be up to the task? If GNSS timings are lost can the system generate it's own timing masters in a resilient way? Sorry I'm probably getting too technical, I tend to go down rabbit holes till I cant go any farther then I get bored and find something new to fixate on. Started looking up how starlink worked wich lead me to ua-cam.com/video/ytBmoL2wZLw/v-deo.html then I found your series. Musk first described the terminal (subscriber module) as a pizza box sized antenna, which I assumed meant phased array or mimo array but now he's calling it a ufo on a stick with servo motors to point it, which I'm thinking is a parabola. I hope it's not a parabola because that would need quite a bit of jitter when it switches satellites. Any reason why a mimo array need to be physically aimed? The satellites only have a maximum beam angle of 25° so I presume the terminals would not need much more than that either.
Synchronization is indeed one of the challenges. The radio stripes architecture described in the following video can deal with the synchronization between some of the antennas: ua-cam.com/video/ZepQAw21HfA/v-deo.html Each antenna unit in the stripe will have a local oscillator that is kept synchronized using a sync-signal that sent via the fronthaul. Synchronization between different radio stripes might be more challenging to achieve and will require some engineering effort. The sensitivity to synchronization errors depends on what carrier frequency is used; higher frequencies are more sensitive. By the way, you don't need an exact phase synchronization to make two signals add constructively at a given location. What is sensitive to synchronization is if you want two signals to cancel each other at a given location.
How can we enhance spectral efficiency by using power control mechanisms? when mobility scenarios the master Ap continiously changes how we can handle that for worst case scenarios?
The transmit powers can be optimized to maximize the sum spectral efficiency or some other metric, by finding the right balance between strong signals and low interference. The mathematical details are found in our book “Foundations of User-Centric Cell-Free Massive MIMO”. The selection of the master AP can be changed every time the user moves sufficiently much. This is also mentioned in the book.
Sorry about the sound issues, there seem to have been some glitch with the microphone, but fortunately it only appears at a few places. Cell-free Massive MIMO is particularly for stadiums, streets, and places with many users or where it is hard to provide sufficient network quality with other methods. There will soon be another video about that.
As per one of your previous videos, to have a narrow beam we must use more antennas and massive mimo by having degree of freedom in number of antennas can provide for our need. I wanna to know in distributed massive mimo and especially in cell free massive mimo that APs usually equiped with single antenna, how can we do beam forming and use spatial multiplexing advantage?
You will have the same degrees of freedom in cell-free and cellular systems if the same number of antennas are used. However, the key difference is that the radiation patterns look different. We are not forming angular beams in cell-free but constructive and destructive interference patterns that are more complicated.
No, 5G is using OFDM, not FM. The data rate is basically proportional to the bandwidth, which is explained in this video: ua-cam.com/video/kP_FhaclHPg/v-deo.html
Basilica Feng, not that I’m aware of. This is a type of resource allocation and such functionality is normally not standardized. But maybe there has been some work in the context of coordinated multipoint.
@@WirelessFuture So, what's the major difference between cell free network and CoMP. I notice that in an old paper (ieeexplore.ieee.org/document/6736745), there are even more names addressed, such as liquid cells, soft cells, and phantom cells, etc. I also don't know what's the difference between cell free network and these names.
In that way, the dynamic cooperation clustering algorithm proposed here is not applicable if the number of users K larger than the number of APs L. As each AP will serve at most tau_p users to avoid pilot contamination. So if we have 10 APs with each AP 10 pilot sequences (tau_p) and 200 users, each AP will serve 10 users at most. What shall other users do?
Cell-free Massive MIMO is meant to be a setup with L>>K, that is what the word “massive” refers to. In this case we won’t get the issue that you refer to. If the network would become overloaded in the way that you describe, then time-frequency scheduling will have to be used to reduce the number of active users per coherence block.
@@marwakhaledyoussef2782 That’s a fair point. What we meant is that the expressions don’t rely on asymptotic properties. But that doesn’t mean that the performance will always be good. A good selection of the clusters is needed. Footnote 5 discusses that it might happen that the Master AP run out of pilots.
Here are two blog posts about it: ma-mimo.ellintech.se/2017/01/25/channel-hardening-makes-fading-channels-behave-as-deterministic/ ma-mimo.ellintech.se/2018/04/23/estimating-channels-under-channel-hardening/ If they don’t answer your doubts, you can ask specific questions in the comments field.
It is the digital link between the site where the antenna/radio equipment is and where the baseband computations are done. Here it is a link between what we call the access point and the central processing unit.
You explain in a nice way, thank you
Emerging field. Enthusiastic
Good night.
Maybe the concepts are different. But I have seen something similar, in Lampsite (small cells from provider X), where they could define several pRRU (pico Remote Radio Unit) related to the same cell. The connection is with PoE.
Similar concepts have been discussed for 20 years. There were a category of collaborative methods between access points (remote radio units) in the 4G era called “coordinated multipoint”. What we talk about in the context of cell-free massive MIMO is the method called “joint transmission/reception”. Huawei has a product that they call LampSite which seems to provide similar features for indoor use. I am not familiar with what kind of algorithms they are using and whether it matches with cell-free massive MIMO, or is more simplified.
Hi Professor, thank you for the informative video. I have a question regarding the Uplink Spectral Efficiency (SE): Why does the scenario where all APs serve all UEs yield better results than the DCC (Dynamic Cooperation Clustering) version? I would think that having all APs serve all UEs might cause more interference. It seems that the DCC version could potentially achieve better SE by reducing interference through selective AP-UE connections. Could you please clarify this? Thank you!
If the uplink processing is not designed to deal with interference, then some APs can cause more harm than good, in the way you described. You can find numerical examples of this in the paper “Cell-Free Massive MIMO: User-Centric Approach”. However, that cannot happen if the uplink processing is optimally designed. When all APs serve all users, you have more options available so you can choose to disregard the received signal if it contains more interference than useful signals. Since you have more information available when you consider the received signals from all APs instead of only a subset of them, you can implement the same receiver processing and more, which is why the optimal solution cannot be worse.
Good explanation! A small question, at around 8:15 , you mentioned that in TDD mode, the user send pilots and base station estimate the channel. Then in the downlink, should the base station send pilots and user estimate the channel? ❤ Thanks!
This is a good question that I tried to answer in a blog post a few years ago: ma-mimo.ellintech.se/2018/11/02/when-are-downlink-pilots-needed/
From a theoretical standpoint, it is sufficient to send pilots in one direction and then use it to make the effective channel predictable in the other direction.
However, in practice, we often send pilots in both directions to deal with hardware impairments. Insufficient calibration between the hardware components used in uplink and downlink will lead to slightly different channels. This is particularly the case with phase drifts that are added with different signs in the uplink and downlink.
@@WirelessFuture Thanks for your explanation! In this blog, it is assumed that uses are equipped with single antenna and downlink pilots are not needed theoretically. For the case where multiple antennas are equipped, should we send downlink pilots? I think users must know h to combine signals ( MRC et al.) and hence downlink pilots are needed. Is that true? Additionally, if downlink pilots are needed, the number of downlink pilots is proportional to the number of antennas at the base station M, which may incur high signaling cost in massive MIMO, right?
You are right that downlink pilots will be needed whenever the user device has multiple antennas. In that case, one should transmit one pilot per downlink beam, and beamform/precode it as one will do what the downlink signal. The number of downlink pilot is then equal to the number of beams, which is typically much lower than the number of base station antennas. So the signaling code is not overly high - typically the same as the uplink pilot signaling.
@@WirelessFuture clear explanation, thanks for ur patient response!
pl discuss Energy efficiency and computation complexity in Cell free Missive MIMO
Great Illustration! Regarding the pilot channels, I have a question: the Master AP listens and tries to find out which pilots are received with the least interference, so in other words you are pointing out to the pilot channel which is not chosen " that much" by the "slaves" AP's to serve certain UE's in which they form a Master AP with respect to them ?
Mohamad Mestou Kirdi Yes, that is a reasonable way to explain it.
@@WirelessFuture But if may I ask, do you think that the Master AP listening to the pilot signals transmitted from other AP's on a certain resource set would guarantee that the next transmission instance of this resource set would lead to similar interference level at the time where the chosen pilot shall be transmitted? Kindly correct me if I'm mistaken, the resource sets can be changed from one instance to another, when it comes to the number of RS they hold, which in turn depends on the number of UE's which the AP is trying to serve, which is also varying(The number of UE's under one Master AP) ? Isn't it better to centralise the pilot status of the AP in each cluster at the centralised controller which can eliminate this kind of ambiguity ?
Do you only assume one CPU? What happens then if there is more than one CPU? In one of your publications, you mentioned that the real case ( in practice) is a system with multiple CPUs.
We are normally considering that there is one CPU with access to all information that the APs are sharing. However, it doesn’t have to be one physical entity. The CPU tasks could be distributed between multiple servers. We mention this briefly in Chapter 1 of the book arxiv.org/pdf/2108.02541.pdf
One can also optimize the allocation of the CPU tasks over physical processing units, as in this paper: arxiv.org/pdf/2301.06166.pdf
Hello professor. I have a question that is bothering me lately. When I search the literature I find that 95% of the works use only spectral efficiency/rate as a performance metric and nobody consider the BER as a performance metric. No doubt the capacity say a lot about the performance but ultimately the bits need to be correctly decoded at the receiver. In that case why is it that nobody is giving any attention to improve BER? I'm working on CF-mMIMO-OTFS and even the first paper on CF-mMIMO-OTFS has mentioned that the BER is an important metric but they left it as a future work. I tried simulating CF-mMIMO-OTFS and I found that the BER is good as long as the number of users are restricted to 2. But as soon as I increase the number of users to more than 2, the interference power increases significantly because of which the BER degrades significantly as well. Can you please give your insights on how to improve this? Perhaps NOMA can help? I'm not sure. Thanks in advance
This is a great question that I have received many times. Maximizing capacity and minimizing BER often go hand-in-hand from an overall design perspective (i.e., how to point beams). However, in practice you often pick a modulation/coding scheme that gives an acceptably small BER, so it is not of a constant than a variable.
I elaborate more on this in a new video (please feel free to ask follow up questions related to it):
Bit Error Ratio or Channel Capacity: Which metric should I use? [Frequently Asked Questions]
ua-cam.com/video/sK537pnqjWM/v-deo.html
@@WirelessFuture Dear professor, thank you for responding it with a video. I agree with your point but then again it is true provided the detection is done correctly. What I do not fairly agree with some of the works is especially when they show capacity plots for SIC-based MMSE detection in CF-mMIMO when in reality SIC never works in cases with more than 2 users. When we can't even do detection this way and can't get a good BER, isn't it redundant to present the capacity plots in this case? I feel both capacity as well as BER together should be considered as a performance metric.
There are no decoding error when considering the capacity - it is achieved with zero BER (and even zero error in the absolute sense) as the block length goes to infinity.
However, in practice, there will certainly be decoding errors and SIC makes the situation worse through error propagation.
It is not the BER but the block error probability that matters in practice: What is the probability that a coded data block is decoded correctly? If it is decoded correctly, we have BER=0. If it is decoded in error, it doesn’t matter what the BER is because you anyway need to retransmit the data.
I imagine for this to work the each site has to be synchronized to an extremely high degree. I would presume GNSS would be the standard go to but several nations have indicated that they are more than willing to turn the systems off if they feel the need to do so additionally these systems are susceptible to jamming and spoofing. They also need high bandwidth low latency site to site traffic if they are to do tricks such as computational interference subtraction, I think this would be orders of magnitude more than the actual data you are interested in, you probably need the raw IQ data. How long is a 5g frame? Will each site need to be meshed together with fiber or will microwave backhauls be up to the task? If GNSS timings are lost can the system generate it's own timing masters in a resilient way?
Sorry I'm probably getting too technical, I tend to go down rabbit holes till I cant go any farther then I get bored and find something new to fixate on. Started looking up how starlink worked wich lead me to ua-cam.com/video/ytBmoL2wZLw/v-deo.html then I found your series.
Musk first described the terminal (subscriber module) as a pizza box sized antenna, which I assumed meant phased array or mimo array but now he's calling it a ufo on a stick with servo motors to point it, which I'm thinking is a parabola. I hope it's not a parabola because that would need quite a bit of jitter when it switches satellites. Any reason why a mimo array need to be physically aimed? The satellites only have a maximum beam angle of 25° so I presume the terminals would not need much more than that either.
Synchronization is indeed one of the challenges. The radio stripes architecture described in the following video can deal with the synchronization between some of the antennas: ua-cam.com/video/ZepQAw21HfA/v-deo.html
Each antenna unit in the stripe will have a local oscillator that is kept synchronized using a sync-signal that sent via the fronthaul. Synchronization between different radio stripes might be more challenging to achieve and will require some engineering effort. The sensitivity to synchronization errors depends on what carrier frequency is used; higher frequencies are more sensitive. By the way, you don't need an exact phase synchronization to make two signals add constructively at a given location. What is sensitive to synchronization is if you want two signals to cancel each other at a given location.
How can we enhance spectral efficiency by using power control mechanisms? when mobility scenarios the master Ap continiously changes how we can handle that for worst case scenarios?
The transmit powers can be optimized to maximize the sum spectral efficiency or some other metric, by finding the right balance between strong signals and low interference. The mathematical details are found in our book “Foundations of User-Centric Cell-Free Massive MIMO”. The selection of the master AP can be changed every time the user moves sufficiently much. This is also mentioned in the book.
Nice Video. Some Problems with the Sound. Is it more fore macro solutions or for speciale solutions like a Stadium or a Streat.
Sorry about the sound issues, there seem to have been some glitch with the microphone, but fortunately it only appears at a few places. Cell-free Massive MIMO is particularly for stadiums, streets, and places with many users or where it is hard to provide sufficient network quality with other methods. There will soon be another video about that.
As per one of your previous videos, to have a narrow beam we must use more antennas and massive mimo by having degree of freedom in number of antennas can provide for our need. I wanna to know in distributed massive mimo and especially in cell free massive mimo that APs usually equiped with single antenna, how can we do beam forming and use spatial multiplexing advantage?
You will have the same degrees of freedom in cell-free and cellular systems if the same number of antennas are used. However, the key difference is that the radiation patterns look different. We are not forming angular beams in cell-free but constructive and destructive interference patterns that are more complicated.
Please explain cell free massive MIMO versus small cell complete paper
i would like to know if 5g still uses FM why does increasing frequency leads to more data rate ? how it really changes frequency to data ?
No, 5G is using OFDM, not FM. The data rate is basically proportional to the bandwidth, which is explained in this video: ua-cam.com/video/kP_FhaclHPg/v-deo.html
Thank you so much for the presentation. Are the codes for simulations available in git for this?
atchut ram Not yet, but when the journal version has been published, we will upload the code.
@@WirelessFuture Thank you.
@@WirelessFuture Is it now available on git? Very interested to further study on ur work.
@@huynjinful yes, now the code is available: github.com/emilbjornson/scalable-cell-free
Are there any Study Item or Work Item in 3gpp who are working on this dynamic cooperation clusters?
Basilica Feng, not that I’m aware of. This is a type of resource allocation and such functionality is normally not standardized. But maybe there has been some work in the context of coordinated multipoint.
@@WirelessFuture So, what's the major difference between cell free network and CoMP. I notice that in an old paper (ieeexplore.ieee.org/document/6736745), there are even more names addressed, such as liquid cells, soft cells, and phantom cells, etc. I also don't know what's the difference between cell free network and these names.
In that way, the dynamic cooperation clustering algorithm proposed here is not applicable if the number of users K larger than the number of APs L. As each AP will serve at most tau_p users to avoid pilot contamination. So if we have 10 APs with each AP 10 pilot sequences (tau_p) and 200 users, each AP will serve 10 users at most. What shall other users do?
Cell-free Massive MIMO is meant to be a setup with L>>K, that is what the word “massive” refers to. In this case we won’t get the issue that you refer to.
If the network would become overloaded in the way that you describe, then time-frequency scheduling will have to be used to reduce the number of active users per coherence block.
@@WirelessFuture thank you. But why it is mentioned in the paper that it holds for any value ?
@@marwakhaledyoussef2782 That’s a fair point. What we meant is that the expressions don’t rely on asymptotic properties. But that doesn’t mean that the performance will always be good. A good selection of the clusters is needed. Footnote 5 discusses that it might happen that the Master AP run out of pilots.
Whenever learning about Cell Free Massive MIMO, I come across a term called "Channel Hardening". Could you please explain what it actually means?
Here are two blog posts about it:
ma-mimo.ellintech.se/2017/01/25/channel-hardening-makes-fading-channels-behave-as-deterministic/
ma-mimo.ellintech.se/2018/04/23/estimating-channels-under-channel-hardening/
If they don’t answer your doubts, you can ask specific questions in the comments field.
@@WirelessFuture Thanks a lot, i was mistaken this whole time that hardening is a shortcoming, turns out that it is a desirable property. Got it now.
very informative lecture, thanks for sharing, is there any way to access your slide
The slides can be downloaded here: github.com/emilbjornson/presentation_slides/blob/master/cellfree-pimrc2019.pdf
@@WirelessFuture Thank you so much
What is Fronthaul link?
It is the digital link between the site where the antenna/radio equipment is and where the baseband computations are done. Here it is a link between what we call the access point and the central processing unit.