Lecture 51 : 5G CSI-RS – part II

Welcome back, so we have been discussing CSIRS okay and what we discussed in the last class was we discussed CSIRS up to 8 ports and we will finish discussing the CSIRS today and we will learn about MIMO transmission okay. So, if you recall we were discussing MIMO okay and then we digressed from MIMO and started learning this different kind of CSIRS, different kind of reference signals okay and one primary reason for this different port reference signal is MIMO okay right otherwise you do not need different port transmission okay. Now after learning in detail about these different kind of reference signal with different number of ports you will be able to appreciate MIMO better that is why it is implemented the way it is okay. So, this is the agenda of today's class okay, so this is a quick summary of what we learned in the last class and where we stopped. So, here we are learning it for 8 port CSIRS okay and these 8 port CSIRS are created using CDM, FDM and TDM configuration example 1 is basically 2 CDM, 2 port CDM and then combined with 4 FDM okay, 2 ports and then 8 ports are created using FDM okay example 2, 2 ports in CDM and then rest of the 4 ports are divided amongst FDM and TDM right okay, 2 CDM then this is 2 FDM and then replicated to give you again 2 TDM to give 4 different ports okay. This is the example of 4 CDM combined with 2 FDM okay, 4 CDM over this and replicated over 4 using FDM okay and then this has a density of 1 resource element per antenna port per PRP okay, 8 resource elements per PRP are using to create 8 ports okay, so this is the density okay right. Now CSIRS is configurable such that it may occupy a part or a whole of bandwidth okay because estimate channel needs to be estimated over all the relevant PRPs okay and the example which we have taken it is only for a per PRP, so you see the resource elements being used for CSIRS okay, so if it is 8 port, so you use 8 resource element per PRP, if it is 32 ports we are using 32 resource element per PRP, as I said it is a humongous requirement okay right, so doing it over entire PRP or entire bandwidth part it is a costly affair but that is what you will have to do okay. So, and this is another fact the starting PRP and number of PRPs of CSIRS are configurable with a granularity of 4 PRPs okay, so if you have to like, so this is constraint is that minimum CSIRS bandwidth is 24 PRPs, so you need to estimate over a minimum of 24 PRPs, after that you increase it with a granularity of 4 PRP okay and you can with that you can go over entire bandwidth part also okay right. CSIRS as we have seen okay as a density of 1 Re per antenna port per PRP right, that is what these are the examples we took okay, we will look at that and but the density can be increased that is the point I made in the beginning okay, can be configured for 3 resource element per antenna port per PRP or it can be reduced 0.5 resource element per antenna port per PRP, either it can be increased or reduced depending on how your channel is changing okay right. If the channel does not change across let us say frequency right, so what will you do? In one PRP, you will transmit another PRP, you do not transmit CSIRS, so the density is reduced correct okay right, if the channel is changing, so what you do? Increase the density right, you make it you can actually make it up to 3 resource elements okay. So, for each port you can increase or like you can basically increase the number of resource element over which these pilots are being transmitted right okay, just to like how will you increase the density over here okay, 8 port transmit 8 port, let us say you have to make density as 2, one you are transmitting here, one you are transmitting at this part, the whole structure you replicate over here right okay. For some reason or you transmit over here, then you basically interpolate right and one then you replicate it over here, it becomes 3 RE per antenna port per PRP, that is the point I am making okay, when user is moving fast, you want to estimate then one here, one here, another here, you have the link flexibility of transmitting it like that okay, so right that is the point okay. Now having studied this structure okay, this that was mostly like how they are designed okay, how frequently they are transmitted, that is what it is the time domain structure and it is similar to SRS okay right, why it is similar to SRS? Because what is CSIRS? CSIRS is the downlink counterpart of SRS right okay, they are not transmitted with data correct okay, so you can configure them in 3 different ways and those 3 different ways we discussed in context of SRS okay, periodic, aperiodic and semi-persistent, you will see the same structure okay right. So, for periodic CSIRS transmission, again RRC configures it, goes on right, that is the periodic CSIRS transmission okay, then semi-persistent is similar to this, but MAC has to get into picture okay, it will be configured again by RRC, but it will be activated by the control element sent by the MAC okay and then again will be deactivated by the control element sent by the MAC layer right okay. Then what is this? This is the aperiodic okay correct, so this is basically what are we saying here that this DCI, this is enabled by the DCI okay, configured by RRC, but enabled by or disabled by DCI okay, so enabled or disabled by DCI means that DCI will say please transmit CSIRS for this much duration that is all okay. Again if it has to be CSIRS has to be basically triggered, then DCI has to be sent again which will tell that the CSIRS has been transmitted okay. This one it is basically what is it telling okay, then we did not discuss in context of SRS what is this same slot receptions, what is this DCI telling that if the slot offset is 0, it means that CSIRS is being transmitted in the same slot okay, if the slot offset is 6, it will be transmitted after 6 slots, that is what it means okay right, so that period is configured for CSIRS, we will see that okay. But this is broadly the idea right and I am not going to in detail because we have discussed this in context of SRS anyway okay right, so this is similar to SRS and both of them are transmitted independently of data, primarily to collect the channel state information okay, one is for downlink, other one is for uplink okay. So, for periodic and semi-persistent CSIRS transmission okay, a device assumes that a CSIRS transmission occurs every nth slot okay, is the period okay, so what is the period between the CSIRS transmission okay, what are we saying every nth slot, where n can be between 400 to 640, these are the number of, this can be the period, so and is also configured with a specific slot offset for the CSIRS transmission, I will show you that okay, the picture will become clear okay. So, what is, how is it configured, what is this picture showing, so here the CSI period is 5 slot okay, CSI is coming, RS is coming every 5th slot, that is what I said, n can be between 4 to 640, so this is n right okay and what do we mean by offset, this here only you are like this, this is let us say this is what is that, you are configuring over here okay and the slot offset is in the same slot you are sending it, so the offset is basically 0 okay, period is 5 right okay, what about this, here again the period is 5 slots right, but you configured it here, this offset is 3, that is all okay right. So, see this is same period but different offset okay, here same offset but different period right okay, so here the period is 10 slots, but you are configuring it here with an offset of 3 okay right, so that is the configuration of CSIRS okay. Now again CSIRS as we understood for SRS that they are basically QPSK modulated okay and each that is what this bullet is telling, each CSIRS resource element is QPSK modulated which is generated as follows, this is the formula we saw okay right, so this is the CSIRS QPSK modulated, what is C again, remember this, scrambler okay, length 31 goal sequence we know how it is done. So, I am not going to go into details of this, but this is how the CSIRS will be QPSK modulated value for each OFDM symbol of a slot index by 0 to 30, this is how you initialize this scrambler okay, I am only telling you briefly how it is done, exact formula okay, you have to look into the standard okay, that is how it is done okay right. So, what have we learnt, where are we, we have finished discussing CSIRS okay right, key points over here is are that CSIRS codes have multiple use cases okay, we have at least seen 2, one is for estimating wherein the pre-coded channel, UE estimates the pre-coded channel okay, using the pre-coded CSIRS sent by the base station right okay, so there the number of code requirement is small okay, because pre-coded channel is compressed. Then second one is un-pre-coded channel that is typically used for FDD system, pre-coded channel is used for TDD systems okay, then this un-pre-coded channel because it is not compressed, you have to use larger number of CSI port which is dependent on the number of transmit antennas okay, in practice okay right, there is a caveat there also. So, because if you look at it, you are defining 32 antennas, maybe you are defining 32 ports, how many, what is the dimension of compressed channel, uncompressed channel you can estimate right, it is basically H should be Nu cross NT right, that is what you can estimate, that essentially means that NT can be a maximum of 32 right, but base stations are being designed with 64 antennas also okay, how do you estimate the channel of that okay, I am not getting into the details, because I can briefly tell you, so that you have a some picture of it, see like our system okay is designed for a 64 antennas, but CSIRS is only 32 port okay. Now base station compresses the 64 antennas channel to a 32 antenna, 32 port channel, how does it uses, how does it do? It uses a fixed pre-coder, which does not depend on the channel okay right, so those pre-coders can be designed without channel information okay right, see I can design the system for 128 antennas also okay right, that standard does not perform any, standard does not restrict me to design or install the number of higher number of antennas, but what it restricts is that I can estimate antenna channel only for 32 ports okay right, so if you are designing a 64 antenna system okay, so you use a fixed pre-coder to compress a 64 antenna channel to a 32 antenna channel okay, you can design a pre-coder right okay, which does not depend on the channel information okay, so that is how it is typically done, I am just leaving you with that okay, and you should not think that with a 32 port CSIRS, you can only design a 32 antenna system, no, you can design a higher antenna system, but you will have to use something on a fixed pre-coder okay right. So this is the CSIRS okay, now after discussing this what we are doing here is, we will see how MIMO is implemented in the standard okay right, so again before we go there CSIRS for TDD, you need a lower number of ports, but again with a caveat that you should have done the TDD calibration okay right, if you have not, if you are not able to do a TDD calibration, then you again switch to un-pre-coded CSIRS transmissions, so that UE can estimate the downlink channel, UE can decide the pre-coder okay right, and then feed it back okay, we will learn that also, how that process is done okay, so this process of UE estimating the channel, feeding it back is commonly done for FDD okay, but it could be done for TDD also, if you are not enabled, if you do not know how to do TDD calibration okay, I will stop with that, now we will start discussing how MIMO is implemented in the standard okay. So, I will recapture it based on the how much time we have, then we will continue, so this is our system model right okay, here what we are doing here is that this is the pre-coder, this is a pre-coded transmission okay, NL number of layers okay, V is the pre-coder, either the base station knows it or UE feeds it back correct okay, now so what will this pre-coder do? It will take in NL layers and then it will map it to NT antennas correct okay, then you transmit the signal over this channel, you receive it over NR antennas and then multiply it with the receiver and then you get these NL layer backs right, remember this picture okay. So, what is this? Y is equal to HX tilde plus N, what is X tilde? X tilde is pre-coded signal, V of X and then it becomes H tilde of X plus N correct okay and the receiver you multiply it with W Hermitian Y right, so we are very compatible with this, for TDD system started till now, G node will design the pre-coder V and decides the rank, rank is basically number of layers, if it knows the channel, it knows the rank okay, the dimension of V is NT cross NL, input is NL layers and the output is NT cross 1 vector, what do you get? What is this? What is the output of this pre-coder V of X? What is this? This is a vector right and what is the dimension of this vector? NT cross 1 correct okay right, rank is equal to the number of layers, equivalently the number of parallel streams which are being transmitted on the same resource element okay fine, this is only the recapturing okay and UE can employ any receiver right, we discussed one receiver, what was that? SVD base okay right and there are other different kind of receivers which are commonly used okay, we are not going to the details of this but zero forcing also we have seen, what will zero forcing do? How will the zero forcing receiver look like? Instead of like WH, what will you do? H tilde inverse right, then also you will if you do H tilde inverse, if you multiply Y with H tilde inverse, what will you get? Again, you will get X but this noise will be different, that will degrade the performance okay right. There are other different kind of receivers MMSE, MMSE, SIC okay, MMSE is minimum mean square, this is minimum mean square, successive interference cancellation, I am introducing you to this because the way this MIMO architecture is defined in the standard, it uses this idea also okay, so I am just telling you that okay, then there is maximum likelihood receiver, we briefly touched upon it right, maximum likelihood in the beginning of the course somewhere okay right, so this is our system model okay, fine right. This is what we have learnt, now what we are saying here, do not look at this okay, I will tell you what is happening over here, so the configuration what we are assuming here is the base station has NT is equal to 64 physical transmit antennas, physical antennas okay right, so H and that it is NU cross NT, this is that NT okay right, so what are we saying here, so if it is if NL is up to 4 layers okay, what do we mean in the channel rank is 4, base station knows it some okay, whether it is a TDD or FDD system okay right, we assume that base station knows it now okay right okay. So what happens in this case, how MIMO is implemented, that is what you see phi receives one transport block, if it is MIMO transmission is up to 4 layers okay, MAC will send only one transport block okay, they send only one transport block, fine okay, so what happens okay, so let us look at this, this is one transport block correct okay, this transport block undergoes the processing, we understand this processing and I have taught you till here right okay. This is what I am saying here, at the end of this what you have one modulated code word complete right okay, segmented all that is done, concatenated, scramble, modulated, at the end of it you have single code word right, concatenated code word okay, now the layer is 4 okay, if you remember the way the transport block size was calculated, number of layers also play a role right okay right, if number of layer is 1, the number of resource block let us say are 10 okay, correct, what will happen okay right. So, the transport block size if the number of layer is 1, let us say it is 100, just for example okay, when compared with NL is equal to 4, MAC will send a transport size of 400 okay, even though the same PRB, over 10 PRBs you need to transmit okay, why? Because now with 4 layers, you can transmit 4 times the data, amount of data which you can send with NL is equal to 1 layer, over the same number of PRBs, that is what we have learnt in all this MIMO processing right, correct okay, so the number of the parallel layers which you can transmit over same number of PRBs right okay, that is what is MIMO okay. So, what are we saying here, 5 processes this transport block to give one modulated code block, till here there is no distinction okay, now after that you see this layer mapping block is coming okay, layer mapping block takes these modulated symbols which were denoted as D okay right, Q is this index, what is the index? What is this Q denoting is a transport block index okay, Q is 0 okay, so it will take from the single code word as input, layer mapper and then what will it do, it will demultiplex this DQI into 4 layers, if NL is 4, correct okay, you get this, what is happening okay, one transport block, the size of transport block was dependent on the number of layers of course okay, right, if it is NL is equal to 1, then this layer mapping block will only be a pass through, correct okay, right, so if it is 4 layers, this you get a 4 layer okay, then there is a pre-coder, what will this pre-coder do, we have learnt in great detail, it will take 4 layers and map it to 64, what is the dimension of V, 64 cross 4, 4 input, 64 output, because what is 64, 64 is the number of antennas, correct okay, right, so that is what is happening, so it pre-codes, now how many such resource block mappings you will have to do, we have learnt this till we have avoided this, what was happening, resource block mapping single and then CP addition, correct okay, right, so we were creating only one resource block mapping, one IFFT, one CP, now how many such processes you will have to do, you have 64 antenna, so you take this pre-coded signal, what is the output of, what is the length of this pre-coded signal, each one of them is a 64 cross 1 vector, right, if MIMO was not being done, what was it, 1 cross 1, we are taking 1, you are doing, you are mapping over one resource block grid, taking IFFT, CP, then doing over one RF chain, right, okay, now what, depending on the number of antenna which you are assuming to be 64, you take this first element of this pre-coded vector, map it on to first resource element grid, pass it on to the DACs in the RF chain, take the second one, take the third one, you have to do it for 64, look at the complexity of doing it, right, this is basically the hardware complexity, right and the following, you will have 64 RF chains, you will have 64 into 2 ADCs and DACs because for each chain you need 2 ADCs and 2 DACs, correct, okay, so that is what is happening, okay, so this is the complexity of doing MIMO, okay, in terms of hardware, because this is simple, this is simple, yes, to get this you know what is required, right, okay, to get this pre-coder information, all that CSIRS whatever needs to be done, it has to be done, then only it can be enabled, okay, that is why I introduced you to that, it is only one block, okay, but the underlying complexity of getting this information is huge, okay, what is happening here, okay, so what is this, the output of this pre-coder I am writing, output is sorry, layer mapper I am writing, okay, is being mapped, you take this, right, I is running from 0, okay, how many layers? 4 layers, okay, right, so this is layer 0 will be 0 symbol, layer 1 will be first symbol, layer 2 will be third symbol, layer 3 will be fourth symbol and then you again go back, there will be 4, 6, 6, 7, 8, 9, 10, 11, that is how you are creating 4 layer from the single transfer block, right, see in this processing, okay, MAC is sending only one transfer block, if the rank is up to 4, 1, 2, 3, 4, okay, single transfer block, then after that number of layers are created at this stage, right, okay, so and that is how this demultiplexing is done, I will run from, I will keep running, I is a running index, okay, so every fourth symbol becomes 1 layer, that is what it means, okay, if the rank is 4, if rank is 2, every second symbol becomes 1 layer, okay, right, so it will be if 2, if rank is 2, it will be 2i, 2i plus 1, this is 0th, this superscript is 0 transfer block because the rank is more than 1, you will get another transfer block, so this index will be 0 and 1, right, okay, so for this 0th transfer block which is, which you single layer gets for a rank of 4 up to a rank of 4, this is what will happen, you understand all the superscripts, okay, right, this is i, the running index, 0 symbol of layer 0, 0 symbol of layer 1, basically the first symbol of this modulated transfer block, 0 symbol of second layer is the third symbol, okay, then the 0 symbol is basically the, this is 0th of 0, 1st of 0, it is the only the index, okay, right, so M layer symbol, okay, what will happen, number of modulated symbols per layer will be how many, if the rank is 4, you are taking those symbols and de-multiplexing, okay, so the number of symbols in each layer will be the total number of symbol divided by the rank, correct, okay, so that is what it is, so in this case, it will be M0 symbol by 4 for 4 layers, okay, right, so and since we considered NT is equal to 64 physical antennas, okay, the chain will have 64 resource block mapping, that is what I said, okay, 64 IFFT you will need to take for each of this, because you are creating 64 resource grids, okay, and you will have CP and rest of the chain also accordingly, okay. So, I think this is what I have for this class, we will continue discussing this, we have how many layers for MIMO transmission for up to 4 layers, okay, right, and why we discussed all these reference signals, okay, just to make you understand, okay, what kind of complexity is involved in enabling MIMO, okay, right, this is not, so this reference signals, the number of ports increases, okay, for MIMO transmission, we have to like understand that even if the MIMO is not there, okay, the SRS, all these transmission need to happen, okay, even if this block is not there, because SRS you need for estimating the uplink channel, okay, when the TDD system, then you can infer the downlink channel, CSIRS you need, so that you can find out the MCS and feed it back, okay, right, even for single layer, DMRS you need to send, okay, because it has to be used by the user to demodulate the actual data, PDSCH data, right, and it should be able to get the instantaneous channel. In all these cases, when MIMO is not there, all those transmissions will be single port, correct, and depending on what is the rank of the MIMO channel, the number of ports of these reference signals is going to increase accordingly, okay, right, and that is where those reference signals, one of the uses to basically design these pre-coders, okay, so that you can use these large number of antenna, which you have, which you have designed, okay, we will stop here, we will continue discussing this, thank you.