Channel State Information Reference Signal (CSI-RS) and Sounding Reference Signal (SRS)
From the series: 5G Tutorial
This video discusses signals in 5G New Radio (NR) that enable channel sounding. Those signals include the channel state information reference signals (CSI-RS) on the downlink and sounding reference signals (SRS) on the uplink. Channel sounding provides channel information used for resource allocation and beamforming. This video explains the time and frequency structure of CSI-RS and introduces zero-power CSI-RS and their role in interference measurement. You’ll learn how CSI is fed back to the gNodeB via codebooks. Then, you’ll learn about the time and frequency structure of SRS and how suitable uplink beamforming is communicated to the UE as a result. Finally, this video walks through the difference between the role of CSI-RS and SRS and the role of DMRS.
Published: 27 Dec 2019
This is a new episode of our series 5G Explained. In this video, we discuss signals that are available in 5G NR to support channel sounding and channel estimation.
First, we will define channel sounding. Then, we will look at channel state information reference signals or CSI-RS, which support channel sounding on the downlink, and sounding reference signals or SRS, which are their equivalent on the uplink.
Much of 5G NR relies on beamforming to provide sufficient SINR at the reception site, in particular for millimeter wave transmission. But beamforming requires knowledge of the propagation channel.
The propagation channel depends on the transmit frequency. Therefore, if uplink and downlink operate on two different frequencies as is the case in FDD, there is no choice but relying on the receiver to communicate information about the channel back to the transmitter. This is the case at the bottom right.
In the case of TDD, where the uplink and downlink share the same transmit frequency, it is possible, on the other hand, to estimate the downlink channel based on measurements on uplink transmission (or the opposite).
Whichever case is applicable, special sounding signals are relied upon for both directions. Those sounding signals are the CSI-RS for downlink and SRS for uplink.
CSI-RS are not a new concept in 5G NR. They have been used in LTE for the same purpose, although the detail of their specification is different in 5G NR.
CSI-RS are transmitted within a bandwidth part, as anything that a UE is expected to listen to. They help with beamforming, interference estimation, as well as coordinated multipoint transmission or CoMP.
As a side note, there is one subtype of CSI-RS called tracking reference signal or TRS, not to be confused with phase tracking reference signals or PTRS, with slightly different goals and characteristics. We will mention it again a little later.
CSI-RS are limited to a bandwidth part, but they are not limited to the subset of the bandwidth part that contains data for the UE- if there is even data transmission. Therefore, they provide information about the channel across possibly the whole bandwidth part, as opposed to DMRS associated with data transmission.
Let us talk about the frequency and time-domain characteristics of CSI-RS, starting with the frequency domain.
There are two main types of CSI-RS in the frequency domain: one and one half. With density 1, CSI-RS are present in every resource block. With half density, they are present in every other resource block.
A third type, very dense in the frequency domain with density three, is called tracking reference signal, as mentioned earlier. This type of CSI-RS is single port, whereas the other CSI-RS can support up to 32 antenna ports for MIMO application.
Here is a concrete view of density one versus density half. As can be seen, with density one, CSI-RS is present in every resource block, whereas CSI-RS is present in one half of resource blocks on the right.
CSI-RS occurrences in the time domain can be scheduled in three different ways: periodic, semi-persistent, and aperiodic.
Periodic transmission is what it sounds like, and the period is between 4 and 640 slots. Semi-persistent is the same as periodic, but transmission can be momentarily suspended.
Aperiodic transmission happens without pre-defined schedule. UE have to be notified of such transmission with downlink control information or DCI, a concept explained in another episode of this 5G Explained series.
So far, we have assumed that those CSI-RS are known symbols sent by the transmitter to help with channel estimation or tracking. This is indeed the case, at least for those regular or non-zero power CSI-RS.
The standard also defines zero-power CSI-RS, which are zero-valued reference signals. The point about those symbols is that the UE is specifically told thereby that the gNodeB is not transmitting anything at those locations in time and frequency. As a result, the UE can assume that any receive power at those locations is due to interference.
This is why we mentioned earlier than CSI-RS can be used for interference measurement.
This is an example of non-zero power and zero-power CSI-RS, and how you can set them up with MathWorks 5G Toolbox.
Here, the non-zero power CSI-RS have a period of five slots with offset 1, meaning that they are present in slots 1,6,11 and so forth. They have a dot five or half density, meaning that they are present in every other resource block. And they are located in symbol 1 of the slot. They are shown in yellow.
The zero-power CSI-RS have a period of 10 slots with offset 1, meaning they are present in slots 1, 11, 21 and so forth. They have a frequency density of 1, which means they are in every resource block, and the symbol location is 6. They are shown in blue.
Now, we want to discuss what the UE does with CSI-RS it receives. The UE can determine the channel estimate, but feeding it back to the gNodeB would take a lot of bandwidth, because the channel varies across the measurement bandwidth.
Therefore, the feedback is not directly the channel estimate but, instead, the index among a set of matrices that would provide the best beamforming. This set of matrices is predefined in the standard. The larger the set, the more accurate the representation, but the costlier the feedback.
Two types of codebooks are defined in the standard: type I is coarser and hence cheaper to signal, whereas type II is more extensive. Type I is considered adequate for single-user transmissions, when beamforming doesn’t have to be quite as accurate.
For multi-user transmission, however, the gNodeB will want to target each user more precisely and may request type II feedback.
Note that, in either case, the gNodeB is not bound to use the suggested beamforming matrix for downlink transmission to this UE.
CSI reports can be carried by the data or control channels, PUSCH or PUCCH, depending on the configuration and the complexity of the report.
The table on this slide can be summarized very simply by stating that the type of report that can be associated with a type of CSI is what makes common sense. For example, an aperiodic CSI-RS can only be reported with an aperiodic report. Or, on the other extreme, reports for a periodic CSI-RS pattern can be of any type: periodic, semi-periodic, or aperiodic, as reports don’t have to be always sent back.
The equivalent of CSI-RS on the uplink is the sounding reference signal. SRS are sent by the UE to aid in channel estimation. They are sent over 1, 2, or 4 antenna ports.
The same mechanism was already used for LTE, but SRS in 5G NR are a lot more flexible:
· The duration for SRS is 1, 2, or 4 symbols as opposed to always 1 in LTE.
· There is a comb structure with 2 or 4 spacing in frequency as opposed to always 2 in LTE.
· The SRS is not necessarily in the last symbol of the slot.
· The bandwidth covered can be anything between 1 and 272 resource blocks.
This slide illustrates the flexibility of SRS in 5G NR. We can see a two-symbol SRS, as Nsymb = 2. The number of symbols can be 1, 2 or 4.
Every fourth subcarrier is used as specified by KTC = 4. There are two possible patterns: every other or every fourth subcarrier, corresponding to KTC = 2 and 4 respectively.
The first symbol occupied by the SRS is specified by the Ioffset value. Ioffset can be between 0 and 5 and, obviously, it must be such that all Nsymb SRS symbols fit within that slot.
Finally, the spectral extent of the SRS can be anywhere from 1 through 272 resource blocks, as per the mSRS parameter.
We have seen how CSI reports are sent back to the gNodeB by the UE.
Correspondingly for the uplink direction, information resulting from SRS reception at the gNodeB can result in scheduling decisions as well as beamforming / precoding selection for the uplink.
As you may remember from the 5G Explained episode about downlink control information or DCI, uplink precoding is one of the pieces of information that gets communicated to the UE via DCI. This information is part of the uplink grant, which comes as a response to a scheduling request from the UE.
As a final remark, one may wonder whether one could use DMRS for channel sounding as opposed to CSI-RS and SRS. There are many reasons why this is not possible.
To name just a few:
· DMRS do not extend beyond the current data transmission allocation.
· They are not available when there is no data transmission.
· Also, DMRS are in support of the current transmission configuration. They don’t enable to look at other options for later scheduling, such as using more antenna ports than in the current transmission (if there even is a current transmission).
This concludes this episode of the 5G Explained series on signals for channel sounding.