Thanks for sharing this excellent content! What happens if we have a distributed antenna architecture such as D-MIMO and all the antennas are synchronized? Would this place the users in the near-field since the inter-element spacing becomes very large in that case?
Yes, since D-MIMO exploits radiative near-field properties even if that terminology is normally not used. There is an illustration of this at end of Chapter 1 of “Foundations of User-Centric Cell-Free Massive MIMO”, where we show how a room with antennas along the walls lead to signal focusing in a small circular area. However, many papers that analyze the radiative near-field consider the implications of having non-planar phase variations, while all the antennas have approximately the same amplitude variations. The Fresnel approximation builds on that assumption. In D-MIMO, the antennas are so far apart that there will also be substantial amplitude variations among the channel from antennas to a user. Hence, many of the analytical results that have been derived in recent years based on the Fresnel approximation are not applicable in D-MIMO.
@@WirelessFuture As a follow up question: Does the Fraunhofer distance d_fa = 2*D^2/\lambda (D is the largest dimension of the antenna array and \lambda the wavelength) depend on the separation between the antenna elements? For instance, is this only valid if the antenna elements are placed uniformly at a separation of \lambda/2? Will d_fa be the same for the following two cases? Case 1: a ULA of 2 m length and antenna elements are placed at a separation of \lambda/2 from 0 m to 2m. Case 2: antenna array having only two antenna elements one at 0m and the other at 2m.
A lot of work on “massive MIMO” focused on making the algorithms universally applicable so they don’t require a particular propagation environment or array configuration to be used. All such results are directly applicable to both near and far field channels. New challenges arise when we want to exploit the specific channel characteristics in the algorithms to improve things like channel estimation. The broadcasting of information might also have to be done differently.
First off, I have now watched several of your videos and learned a lot. However, I have a couple of questions, assuming TDD and beam forming. I understand using the reciprocity-based approach for determining the best estimation. However, does that still work to acquire the signal initially, when the site does not yet know where a phone is and cannot yet establish a beam? Also, do cell sites still use different frequencies for the three sectors, as had been the case for older generations? Or does beam forming provide sufficient isolation for the same frequencies in all sectors?
These are great questions. The TDD reciprocity-based beamforming concept is particularly utilized for users that are connected to the system and have been scheduled for transmission/reception. For initial access, there are typically random access procedures where any user can transmit uplink signals to request to be scheduled. It resembles transmitting a pilot plus a small data block, but with the risk of collision between users. If the base stations is able to decode the message, it can then schedule the user and transmit that information back in the downlink. The reciprocity-based beamforming for data transmission can then begin on the assigned time-frequency resources. In principle, 5G networks can operate with universal frequency-reuse, so all frequencies are used both across cells and cell sectors. As you said, beamforming is instead used to isolate the transmissions in different sectors and cells. I’m not sure if this is used in all practical deployments or not. Here is a quite recent blog post from Ericsson that writes about this: www.ericsson.com/en/reports-and-papers/microwave-outlook/articles/maximizing-capacity-in-spectrum-limited-networks
Excellent video! Thank you for sharing it! In the case of Rayleigh fading (that is, when there is no LoS component between the BS and the UEs), should we also consider the effects of near-field propagation if the UEs are too close to the BS (i.e., a situation where the distance between the UEs and the BS is smaller than the Fraunhofer distance)?
Hello, I am reading your book. I short question about SVD beamforming. I think we need both matrices, U and V of H. One at the receiver in at the transmitter. Because under example 4.25, there is written at the end "Alternatively, the transmitter can generate two independent data streams......and apply the transmit precoding x=U^H*x..... That sounds to me as if you can also work with just U matrix. Do I need both matrices or can I also work with just one matrix. E.g. U? Thanks a lot.
I'm not sure if I understand the question. Many of the theoretical papers that consider near-field communications assume free-space line-of-sight channels. If we want to verify those results using measurements, then we need an anechoic chamber that absorbs all reflections.
@@WirelessFuture yeah the problem is that chambers concept use complex near to far field transformations (complex mathematical algorithms). This is only to get the results at the far field. In this video you highlight that near field communication “is not bad” (let’s call it like that). Hence I wonder why not avoiding those complex algorithms and test only near field comm. Instead, everyone in the industry is motivated about the far field behaviour.
OK, now I understand your concern. A 5G base station might have a 1 m aperture length and operate at a 0.1 m wavelength, so the Fraunhofer distance is 2*1^2/0.1 = 20 m. Since we deploy base stations at elevated places, all users are at distances beyond 20 m, so we will only communicate in the far-field and are only interested in those propagation properties. However, many anechoic chambers are smaller than 20 m, so we must use near-to-far-field transformations to measure the far-field properties. When people talk about "near-field communications", they consider situations where many of the users are located at distances shorter than the Fraunhofer distance. This can happen when the arrays are larger and/or the wavelength is shorter.
Thanks, in your book in chapter you write about the matrix H^T - you call it the opposite direction. Do you mean that direction on the other open half, i.e. the direction which is 180 deg. shifted? Thanks.
Hi, I was reading a paper, "A Tutorial on Near-Field XL-MIMO Communications Towards 6G," particularly Fig.12. It was shown that the near field channel coefficient gives a higher rank (>1) for the specific channel model in (25) in the same paper. However, as you mentioned in the first part of this video, the near-field channel modeling does not provide such gain. Could you please mention what is my misunderstanding here?
This simulation setup considers 128 transmit antennas and 32 receive antennas, so the maximum channel rank is min(128,32) = 32. However, the “effective rank” can be much lower if the propagation environment contains very few propagation paths. The specific channel model in (25) is for free-space line-of-sight, so then there is only one propagation path. At short distances, Figure 12 shows that the channel rank is larger than 1, although there is only one physical path. This is thanks to the spherical curvatures of waves in the near-field. The point I make in the video is that these are not _new_ dimensions that can enable a higher rank than 32. Instead, one near-field path behaves as multiple far-field paths, arriving from a few different angles around the line-of-sight path. The good thing with near-field propagation is that each physical path behaves multiple far-field paths, which can be a benefit if there are very few physical paths. But as the number of physical paths increases, we will eventually get a similar channel rank in both near-field and far-field scenarios, since it is upper bounded by 32.
In your great book you analyze the capacity equation C=B*ld(1+SNR). One can think it goes to Infinity for B->infinity because of the B in front of the ld expression. That is not the case, ok. Is there a special math. theorem that states that we cannot go that way? Perhaps we have B also within that ld term - we must take the limit analysis and cannot say it goes to Infinity. Thanks a lot.
The reason is that the SNR is also bandwidth dependent: SNR = P_r / (B*N_0), where P_r is the total received power and N_0 is the noise power spectral density. Hence, when the bandwidth increases, the SNR reduces because the signal power must be divided over a bigger bandwidth. However, if you also increase the transmit power proportionally to the bandwidth, you achieve a constant SNR and the capacity grows without bound as B->infinity.
Dear Professor, I have watched the broadcast several days ago and have a question. I agree that the near-field communication wont provide additional DoF for single antenna user, but when the number of antenna at user side increases, the rank of channel matrix will increase even with the presence of LoS path only. So would you think that it is the additional DoF? Thank you.
Dear Shicong Liu, for a given array geometry, there is a maximal number of DoFs determined by the sampling theorem (as described in the video). When you consider the communication link between a transmitter and receiver, the maximum channel rank is the minimum of the transmitter’s and the receiver’s DoFs. This is a tighter bound than the classical rank
Excellent as always, Emil! Thanks for sharing.
Thanks for sharing this excellent content! What happens if we have a distributed antenna architecture such as D-MIMO and all the antennas are synchronized? Would this place the users in the near-field since the inter-element spacing becomes very large in that case?
Yes, since D-MIMO exploits radiative near-field properties even if that terminology is normally not used. There is an illustration of this at end of Chapter 1 of “Foundations of User-Centric Cell-Free Massive MIMO”, where we show how a room with antennas along the walls lead to signal focusing in a small circular area.
However, many papers that analyze the radiative near-field consider the implications of having non-planar phase variations, while all the antennas have approximately the same amplitude variations. The Fresnel approximation builds on that assumption. In D-MIMO, the antennas are so far apart that there will also be substantial amplitude variations among the channel from antennas to a user. Hence, many of the analytical results that have been derived in recent years based on the Fresnel approximation are not applicable in D-MIMO.
@@WirelessFuture Thanks a lot, Emil.
@@WirelessFuture As a follow up question: Does the Fraunhofer distance d_fa = 2*D^2/\lambda (D is the largest dimension of the antenna array and \lambda the wavelength) depend on the separation between the antenna elements? For instance, is this only valid if the antenna elements are placed uniformly at a separation of \lambda/2?
Will d_fa be the same for the following two cases?
Case 1: a ULA of 2 m length and antenna elements are placed at a separation of \lambda/2 from 0 m to 2m.
Case 2: antenna array having only two antenna elements one at 0m and the other at 2m.
Good video! So what problem will near field communication bring compared to far field communication? Thanks!
A lot of work on “massive MIMO” focused on making the algorithms universally applicable so they don’t require a particular propagation environment or array configuration to be used. All such results are directly applicable to both near and far field channels.
New challenges arise when we want to exploit the specific channel characteristics in the algorithms to improve things like channel estimation. The broadcasting of information might also have to be done differently.
First off, I have now watched several of your videos and learned a lot. However, I have a couple of questions, assuming TDD and beam forming. I understand using the reciprocity-based approach for determining the best estimation. However, does that still work to acquire the signal initially, when the site does not yet know where a phone is and cannot yet establish a beam? Also, do cell sites still use different frequencies for the three sectors, as had been the case for older generations? Or does beam forming provide sufficient isolation for the same frequencies in all sectors?
These are great questions. The TDD reciprocity-based beamforming concept is particularly utilized for users that are connected to the system and have been scheduled for transmission/reception. For initial access, there are typically random access procedures where any user can transmit uplink signals to request to be scheduled. It resembles transmitting a pilot plus a small data block, but with the risk of collision between users. If the base stations is able to decode the message, it can then schedule the user and transmit that information back in the downlink. The reciprocity-based beamforming for data transmission can then begin on the assigned time-frequency resources.
In principle, 5G networks can operate with universal frequency-reuse, so all frequencies are used both across cells and cell sectors. As you said, beamforming is instead used to isolate the transmissions in different sectors and cells. I’m not sure if this is used in all practical deployments or not. Here is a quite recent blog post from Ericsson that writes about this: www.ericsson.com/en/reports-and-papers/microwave-outlook/articles/maximizing-capacity-in-spectrum-limited-networks
Excellent video! Thank you for sharing it!
In the case of Rayleigh fading (that is, when there is no LoS component between the BS and the UEs), should we also consider the effects of near-field propagation if the UEs are too close to the BS (i.e., a situation where the distance between the UEs and the BS is smaller than the Fraunhofer distance)?
Yes, if the scattering objects are in the radiative near-field of the BS.
@@WirelessFuture Thanks!
Hello, I am reading your book. I short question about SVD beamforming. I think we need both matrices, U and V of H. One at the receiver in at the transmitter.
Because under example 4.25, there is written at the end "Alternatively, the transmitter can generate two independent data streams......and apply the transmit precoding x=U^H*x.....
That sounds to me as if you can also work with just U matrix.
Do I need both matrices or can I also work with just one matrix. E.g. U?
Thanks a lot.
Yes, you use U at the receiver and V at the transmitter. See Figure 3.11.
Just wonder why the concept of chamber exist to test mmWave OTA with complex transformations from near to far filed are needed
I'm not sure if I understand the question. Many of the theoretical papers that consider near-field communications assume free-space line-of-sight channels. If we want to verify those results using measurements, then we need an anechoic chamber that absorbs all reflections.
@@WirelessFuture yeah the problem is that chambers concept use complex near to far field transformations (complex mathematical algorithms). This is only to get the results at the far field. In this video you highlight that near field communication “is not bad” (let’s call it like that). Hence I wonder why not avoiding those complex algorithms and test only near field comm. Instead, everyone in the industry is motivated about the far field behaviour.
OK, now I understand your concern. A 5G base station might have a 1 m aperture length and operate at a 0.1 m wavelength, so the Fraunhofer distance is 2*1^2/0.1 = 20 m. Since we deploy base stations at elevated places, all users are at distances beyond 20 m, so we will only communicate in the far-field and are only interested in those propagation properties. However, many anechoic chambers are smaller than 20 m, so we must use near-to-far-field transformations to measure the far-field properties.
When people talk about "near-field communications", they consider situations where many of the users are located at distances shorter than the Fraunhofer distance. This can happen when the arrays are larger and/or the wavelength is shorter.
@@WirelessFuture thanks for the clarification professor. Looking forward to coming videos from you.
Thanks,
in your book in chapter you write about the matrix H^T - you call it the opposite direction. Do you mean that direction on the other open half, i.e. the direction which is 180 deg. shifted?
Thanks.
No, this means that the transmitter and receiver switch roles.
Hi, I was reading a paper, "A Tutorial on Near-Field XL-MIMO Communications Towards 6G," particularly Fig.12. It was shown that the near field channel coefficient gives a higher rank (>1) for the specific channel model in (25) in the same paper. However, as you mentioned in the first part of this video, the near-field channel modeling does not provide such gain. Could you please mention what is my misunderstanding here?
This simulation setup considers 128 transmit antennas and 32 receive antennas, so the maximum channel rank is min(128,32) = 32. However, the “effective rank” can be much lower if the propagation environment contains very few propagation paths. The specific channel model in (25) is for free-space line-of-sight, so then there is only one propagation path. At short distances, Figure 12 shows that the channel rank is larger than 1, although there is only one physical path. This is thanks to the spherical curvatures of waves in the near-field.
The point I make in the video is that these are not _new_ dimensions that can enable a higher rank than 32. Instead, one near-field path behaves as multiple far-field paths, arriving from a few different angles around the line-of-sight path. The good thing with near-field propagation is that each physical path behaves multiple far-field paths, which can be a benefit if there are very few physical paths. But as the number of physical paths increases, we will eventually get a similar channel rank in both near-field and far-field scenarios, since it is upper bounded by 32.
@@WirelessFuture Thank you so much for clarifying my doubt and making my understanding clear. 🙂
In your great book you analyze the capacity equation C=B*ld(1+SNR). One can think it goes to Infinity for B->infinity because of the B in front of the ld expression. That is not the case, ok. Is there a special math. theorem that states that we cannot go that way?
Perhaps we have B also within that ld term - we must take the limit analysis and cannot say it goes to Infinity. Thanks a lot.
The reason is that the SNR is also bandwidth dependent: SNR = P_r / (B*N_0), where P_r is the total received power and N_0 is the noise power spectral density. Hence, when the bandwidth increases, the SNR reduces because the signal power must be divided over a bigger bandwidth. However, if you also increase the transmit power proportionally to the bandwidth, you achieve a constant SNR and the capacity grows without bound as B->infinity.
Dear Professor, I have watched the broadcast several days ago and have a question. I agree that the near-field communication wont provide additional DoF for single antenna user, but when the number of antenna at user side increases, the rank of channel matrix will increase even with the presence of LoS path only. So would you think that it is the additional DoF? Thank you.
Dear Shicong Liu, for a given array geometry, there is a maximal number of DoFs determined by the sampling theorem (as described in the video). When you consider the communication link between a transmitter and receiver, the maximum channel rank is the minimum of the transmitter’s and the receiver’s DoFs. This is a tighter bound than the classical rank
@@WirelessFuture Thank you for the explanation!