5G/NR - Technical Challenges

 

 

 

NOTE : This is the note that I wrote long before 5G specification is finalized. Some of the technical challenges mentioned here has already been resolved and some of them are still to be improved.

How will you achieve the technology ? what would be the challege ?

Since no technical specification is done for 5G, I cannot talk anything about any technology for 5G that is officially described. What I want to say in this section is just to think of some possible technical factors and challenges for those technical factors. Most of the comments for each factors would sound too pessimistic, but it has always been like this before any new technology come out, but I believe eventually most of these challenges will be overcome or totally new concept that can overcome these obstacles will emerge. This is how all the current technologies have evolved. The purpose of my comments here to give you something to think about.

One thing for sure for 5G would be that the required data throughput for 5G will be much higher (probably tens of Gbps) than 4G. So I want to talk about general technique for increasing the data rate.

Increasing Modulation Depth

Whenever a technology reach a point where it has to jump up the data rate, one of the first step has almost always been "to increase Modulate Depth". The most common evolution path for the modulation Scheme was as follows :

         BPSK -> QPSK(QAM) -> 16 QAM -> 64 QAM

What would be the next step ? Logically it should be 256 QAM. (If it is too much, we may think about 128 QAM)

Would this be possible ? It may be possible if it is wired, it may be possible (possible but not easy even in wired communication). However, it would be very difficult (almost impossible ?) with wireless communication when we think about various wireless factors (AWGN, Fading, Avaliable Power, Dynamic Range, PAPR handling etc). <-- This is the status as of I first write this which is about 2 years ago. Now (as of Oct 2015), 256 QAM in current LTE is already implemented in some chipset and being tested in UE level.

The Ceiling Moved Twice

The guess above is that 256 QAM comes next, with 128 QAM as the fallback. That guess was right, and the ceiling then moved once more. What the specification did not do is push in one direction only.

NR carries 256 QAM from its first release, on the downlink and on the uplink alike. A second modulation and coding scheme table holds it, and mcs-Table set to qam256 selects that table.

Release 17 added 1024 QAM, and the addition is narrower than the name suggests. It applies to the downlink only, and only in FR1. The parameter is mcs-Table-r17 set to qam1024, and the highest entry of that table reaches a spectral efficiency of 9.2578 bits per symbol.

The uplink moved the other way as well, and that is the part no 2013 list would have predicted. 38.211 defines pi/2-BPSK for PUSCH when transform precoding is configured. It sits below QPSK. It exists for the peak to average ratio problem named in the paragraph above, and it improves coverage rather than throughput.

The cost of each step upward is written into the radio requirements. 38.101-2 allows a UE in FR2 an error vector magnitude of 8.0 % at 64 QAM, and only 3.5 % at 256 QAM. So the transmitter has to be roughly twice as clean for one more step. 128 QAM never appeared in any release.

  • 256 QAM was the right guess : it is in NR from the first release, in both directions, and it needs no special conditions.
  • 1024 QAM went one step further, in one place : downlink only, FR1 only, and added in Release 17 rather than at the start.
  • The uplink also gained a lower order : pi/2-BPSK answers the PAPR problem this section raises, and it lowers modulation depth on purpose.
  • Each step tightens the EVM requirement : the FR2 figure falls from 8.0 % to 3.5 % between 64 QAM and 256 QAM.

Increasing Number of Antenna and Spatial Multiplexing factors 

When we reach the maximum modulation depth, another step we think of as the next step to push up the throughput would be to increase the number of antenna and Spatial Multiplexing factors.

The most common Antenna Configuration for multiplexing we use the most commonly as of now (as of May,2013) is 2 x 2 MIMO. The maximum configuration that is specified in 3GPP as part of final goal of LTE advanced is 8 x 8.

Considering the 5G data rate will be much higher than the final stage data rate for LTE advanced, it is highly probable to adopt even more entenna (more complicating Spatial Multiplexing). How many antenna we can use ? How difficult it would be ?

We don't know... but the number of Antenna that SamSung claimed to use for their next generation technology trial (May 2013) was 64 antenna array.

Would this be really feasible/practical ? Look at the mobile phone that you have now and imagine that you have to put a 64 antenna properly spaced for maximize the spatial multiplexing. At least, Antenna and RF designer would not want to even think about it -:)

<-- (As of Oct 2015), it is likely that the Massive MIMO would operate in MU-MIMO (MultiUser MIMO). In this case, eNodeB (Not sure if it will still be called as eNodeB in 5G) will have huge number of Antenna (like 64 or even more than 100) but UE will have much less number of Antenna.

Increasing the operating bandwidth

Bandwidth is the third lever, after modulation depth and spatial multiplexing. Two questions decide its limit. How much contiguous spectrum a licence can cover, and how wide a band the radio hardware can handle at one centre frequency. The paragraphs below take them in that order.

If all the other technique mentioned above is not possible, an alternative we can think of is to increase the channel bandwidth (RF bandwidth).

What is the maximum bandwidth we use for LTE as of now ? (May 2013)

It is 20 Mhz assuming Single Carrier. If we adopt LTE advanced technically available as of now, it can be max 40 Mhz using Carrier Aggregation with 2 carriers. At the final stage of LTE advanced, we may be able to use max 100 Mhz using 5 carriers.

If you think about 100 Mhz for the currently available spectrum (mostly 800~3Ghz), first you would have spectrum license issue. Most of the spectrum is already sold out. Another issue would be to develop various RF components to handle 100 Mhz contiguous spectrum. 100 Mhz single band would be too much fractional bandwidth (operational bandwidth divided by the center frequency). You can think of implementing 100 Mhz using 5 separate 20 Mhz band, but in that case you need to implement 5 separate RF chains for it which will make RF/hardware design so complicated.

<-- (As of Oct 2015). Then you may ask how wide the bandwidth would be in 5G ? We don't know yet, but hearing from chipset vendors , network vendors and test equipment, it is likely to start with under 200 Mhz(e.g, 160 Mhz) or 400 Mhz. Ideal maximum that is being suggested is around 2 Ghz Bandwidth.

How Wide the Carrier Actually Became

The figure named just above is around 2 GHz, and it was a guess made in 2015 about an unfinished specification. It is worth checking against the finished one. Guesses of that kind are rarely accurate.

38.101-2 splits the millimetre wave range in two. FR2-1 runs from 24250 MHz to 52600 MHz, and FR2-2 from 52600 MHz to 71000 MHz. The UE channel bandwidths defined for those ranges are 50, 100, 200, 400, 800, 1600 and 2000 MHz.

So the largest single channel is 2000 MHz wide, and carrier aggregation reaches the same 2000 MHz as an aggregated bandwidth. The 2015 guess was accurate.

FR1 is the other half of the answer, and it is far narrower. It spans 410 MHz to 7125 MHz, and 100 MHz is the widest channel in it. The licensing problem this section describes was therefore never solved at the frequencies the section is about. It was avoided by moving upward, which is what the next section already suspected.

  • 2 GHz exists, above 52.6 GHz : the 2000 MHz channel belongs to FR2-2, and nothing near it is available lower down.
  • FR1 stopped at 100 MHz : the figure treated here as a hard target is still the widest single FR1 channel.
  • Aggregation did not go further than one channel : the aggregated bandwidth in FR2 reaches the same 2000 MHz ceiling.

Using very high frequency spectrum

One of the ways to get around the 'increasing operating bandwidth' issues described above would be to use very high frequency spectrum which has not been licensed out much. For example, if you go to 20 Ghz spectrum, 100 Mhz BW is only 0.5 % fractional bandwidth.. and several hundreds of Mhz BW is less than 5%.

In case of recent SamSung trial (May 2013), it is claimed to use 26 Ghz spectrum.

But one of the biggest problem with this kind of high frequency (millimeter wave) is that it has extremly high path loss and is very vulnerable to various envirenmental/weather factors like building, tree, moisture, rain etc.

Another important problem with this kind of high frequency is that it would be very difficult to get the small sized RF components properly working at this frequency. (For example, RF SAW duplexer, SAW filters which are one of the most commonly used components in the mobile phone currently will not be used in these frequency.)

One area that can be more practical at least as of now (July 2013) would be in 5Ghz range since the chipsets and devices supporting 802.11ac are already emerging in the market even though the performance may not exactly meet the claimed criteria. However, it would be pretty sure that higher and wider spectrum will be considered as well in real 5G mobile communication.

<-- (as of Oct 2015) FCC started promoting 5 different blocks in mmWave range. The lowest one is around 29Ghz and the highest is around 59 Ghz.

Dynamically Configurable Spectrum both in frequency and time domain

Unlike the current communication system, it is expected that multiple radio technology, operators and carrier frequencies would change dymanically even during the signal user service session. If you think of LTE-A Carrier Aggregation, WiFi Offload, it would give you very primitive idea of this. But this kind of combination and dynamic changes will become a default mode of operation in 5G.

Then.. the challenging issue would be how the RF front end of the device handle this kind of situation.

What Replaced the Universal Front End

The worry in the paragraph above is the right one, and the answer was not a front end that handles everything at once. NR narrowed what a UE has to do at any single moment instead. Two mechanisms in the specification follow that approach, and both are visible in a live configuration.

The first is the bandwidth part. A UE is configured with a small set of them inside one carrier, and exactly one is active per direction at any instant. A DCI switches which one that is. So the receiver runs at the active bandwidth rather than at the carrier bandwidth, and a 100 MHz carrier does not force a 100 MHz front end to be awake for a 5 MHz transfer.

The second is the supplementary uplink. One cell is given two uplink carriers on different bands, and the UE transmits on one of them at a time. That is the carrier frequency change within a session this section predicted, and the UE still needs only one uplink chain.

Both answers have the same shape. The hardware was not made universal. The specification was written so that the hardware handles one narrow situation at a time, and switches quickly between them.

  • Switching replaced simultaneity : the difficult case is not two bands at once, and NR arranges for the UE to meet one at a time.
  • The bandwidth part is the frequency domain answer : one active part per direction, switched by DCI, so the front end works at the width in use.
  • The supplementary uplink is the band answer : two uplink carriers on one cell, one of them active.

Massive MIMO

If I am asked to list a couple of critical features that should be accomplished in 5G before anything else, I would list as follows :

  • Extremely high data rate
  • Handling simultaneous (concurrent) users in PHY/MAC in much more numbers than the case in current LTE
  • Handling the path loss which is normally observed in mm Wave (milimeter wave) region

Even though it is not a single solution that can completely achieve all these features, one of the key factor would be Massive MIMO (In case of the prototype that SamSung came up with in May 2013, it used 64 Antenna in 28 Ghz). I stronly recommend you google this topic and study. Following is a couple of introductary material that I found.

How Large the Array Actually Became

The question in this section is how many antennas a system can use. The specification answers a slightly different question, and the difference matters when reading a data sheet. It counts ports, not antenna elements.

38.211 defines a CSI-RS resource with 1 up to 32 ports, and the current release adds table entries for 48, 64 and 128 ports. A port is not an element. One port drives a group of physical elements through fixed weights, so an array of several hundred elements can present 32 ports to a UE.

The prototype named above used 64 elements at 28 GHz. Counted as ports, the specification passed 64 and defines twice as many.

The UE end stayed small, exactly as the 2015 note expected. 38.214 defines the Type I single panel codebooks for one to eight layers, so eight is the ceiling on what one UE receives at once. The asymmetry predicted here is therefore the shape that was standardised. A large array serves the cell, and each Massive MIMO user takes a few layers out of it.

  • Count ports, not elements : a port number in a specification says how many independent streams the array presents, and the element count is an implementation matter.
  • The array grew past the prototype : CSI-RS is defined up to 128 ports, against the 64 elements of the trial described above.
  • Eight layers is the UE ceiling : the single panel codebooks stop there, which is what makes the arrangement multi user rather than single user.

Very Short TTI and Extremly Low Latency

These property can be described in several different layers..but usually when we talk about TTI it usually refer to MAC/PHY property and when we talk about Latency it usually refer to higher layer including IP layer.

Regarding TTI, in WCDMA R99 the most common MAC layer TTI were 10 ms (U-Plane) or 40 ms (C-Plane), in HSPA it become 2ms or 10 ms, in LTE it become 1 ms. I think 1 ms is already a kind of min TTI, but we would need event shorter TTI to achieve the max throughput/latency criteria required for 5G.

Another possibility would be that a couple of additional TTI is introduced and let the system select one of the TTI dynamically or semi-statically depending on situations.

--> (As of Oct 2015) It is not yet determined anything on the TTI length(subframe length) of 5G. But as far as I am hearing, 0.2 ms (200 us) or 0.1 ms(100 us) subframe length is most frequently mentioned.

What Actually Shortened

The guess in the paragraph above is a subframe of 100 or 200 microseconds. The subframe did not change at all. It is still 1 ms in NR, and the shortening happened one level below it.

38.211 lists seven numerologies, with subcarrier spacings of 15, 30, 60, 120, 240, 480 and 960 kHz. The matching slots per subframe are 1, 2, 4, 8, 16, 32 and 64. A slot holds 14 symbols at every one of them.

So the slot length falls as the subcarrier spacing rises. It is 1 ms at 15 kHz, 125 microseconds at 120 kHz, and 15.625 microseconds at 960 kHz. The 100 to 200 microsecond guess sits almost exactly on the 120 kHz slot, which is the usual millimetre wave numerology.

A second mechanism goes below the slot. A downlink allocation of PDSCH mapping type B takes a length of 2 to 13 symbols and may start at any of symbols 0 to 12. It therefore need not fill a slot, and it need not begin at a slot edge. That is where the latency answer comes from, rather than from the numerology alone.

  • The subframe stayed at 1 ms : it is a fixed unit of the frame structure in NR, and nothing about latency changed it.
  • The slot is what shrank : from 1 ms down to 15.625 microseconds, purely as a consequence of the subcarrier spacing.
  • The 2015 guess matched one numerology : 125 microseconds at 120 kHz is inside the range named above.
  • Sub-slot scheduling was needed as well : mapping type B allows 2 to 13 symbols starting anywhere in the slot, and that is what removes the wait for a slot boundary.

Extremely High Sampling Rate ADC/DAC

Even thought nothing specific is formally defined for 5G requirement, I think it will be highly probable that the system bandwith will get extended to at least a couple of hundreds Mhz. As I mentioned above, there will be a lot of issues related to RF (or milimeter wave) but there will also be tough chalenges at baseband level as well. The first question you would have would be "what would be the sampling rate ?". If the system bandwidth at carrier frequency level is a couple of Mhz, the sampling rate at baseband level should be several Mhz BW.

It means that you need ADC which can sample and convert at several Mhz rate. You may think this would not be a big issue since you may have seen various digital oscilloscope which covers even a couple of Gb sampling rate.

Yes.. it is true. This kind of super high frequency ADC is already used in various area especially high end digital oscilloscope. But those high frequency sampling is not done by a single ADC. It is done by multiple ADCs working in parallel and sampling in interleaving pattern. To make this kind of sampling work properly, very complicated hardware design and control algorithm are required.

To make this kind of technology usable for a mobile device which should not be as expensive and bulky as a high end digital oscilloscope, great deal of improvement on ADC/DAC technology is required as well.

See if following materials can give you any insight on this topic. (Think of how you can make this kind of technology usuable on mobile device)

RACH mechanism to handle very large numbers of subscribers

One of the important goal of 5G is to implement a system that can handle very large numbers of subscribers. These subscribers are not only human but also various types of machines and sensors. In this case, the chances would increase a lot of PRACH reaching a network simultaneously and casuing contention. To figure out a way to handle this situation will be an important topics to be researched (especially in MAC layer design).

How NR Scaled the Contention Pool

The concern in this section is contention, once very many devices compete for one channel. NR did not replace contention with something else. It enlarged the pool and shortened the exchange, and both moves are visible in the configuration a cell broadcasts.

Each PRACH occasion carries 64 preambles, and 38.211 fixes that number rather than leaving it to the operator. What an operator does control is how many occasions exist. The field msg1-FDM places one, two, four or eight PRACH occasions side by side in frequency at the same instant.

A second parameter ties those occasions to beams. Its name is ssb-perRACH-OccasionAndCB-PreamblesPerSSB, and it runs from one eighth of an occasion per SSB up to sixteen SSBs sharing one occasion. So a network trades beam count against preambles per beam, and a cell with few beams gives each of them more preambles.

Release 16 then shortened the procedure itself. In the two step version, the preamble and a small payload travel together as msgA, so a successful attempt costs two messages rather than four. Fewer messages per attempt means the same PRACH resource carries more attempts per second.

  • 64 preambles per occasion is fixed : capacity is added by adding occasions, never by adding preambles to one of them.
  • Occasions multiply in frequency : msg1-FDM allows up to eight at the same instant, which is a direct multiplier on contention capacity.
  • Beams and preambles are traded : the SSB to occasion mapping decides how the 64 are divided, and a high beam count leaves fewer per beam.
  • The exchange got shorter, not smarter : two step random access keeps contention and halves the message count.

Choosing the Waveform

Every section above takes the waveform for granted. Between 2013 and 2016 nobody did. OFDM had carried LTE for a decade, and several research groups argued in print that it could not carry 5G. A real competition ran alongside the questions listed on this page, and it is worth recording how it ended.

The case against OFDM has three parts. A rectangular pulse leaks energy outside its own band, so guard bands are wide. The cyclic prefix is pure overhead, and it is repeated every symbol. Frequency offset between users destroys the orthogonality the scheme depends on.

The candidates all addressed the first problem the same way, by shaping the pulse with a filter. FBMC filtered each subcarrier, UFMC filtered groups of them, and GFDM filtered blocks in two dimensions. Each removed some of the leakage, and each cost something in complexity, in latency, or in how well it works with MIMO.

None of them entered the specification. 38.211 has a single baseband generation clause, and it covers every physical channel except PRACH in both directions. It generates cyclic prefix OFDM. The debate was settled by keeping what was already there.

The uplink was given one extra step rather than a different waveform. Transform precoding spreads a slot of symbols across the allocation with a DFT before the OFDM stage, which produces DFT-s-OFDM. It is defined in the uplink clauses only. The downlink chain for PDSCH runs scrambling, then modulation, then layer mapping, and it has no equivalent step at all.

The reason is the same one that produced pi/2-BPSK in the modulation section above. A UE at the cell edge is limited by its power amplifier rather than by its bandwidth, and a lower peak to average ratio turns directly into transmitted power. That argument applies to a handset and not to a base station, which is why the asymmetry exists.

Flexibility, which is what the filtered waveforms promised, came from somewhere else. NR kept one waveform and gave it seven subcarrier spacings instead. The numerology carries the range of use cases that a filter bank was proposed to carry.

  • OFDM was kept : one baseband generation clause covers both directions, and none of the filtered candidates reached the specification.
  • The uplink has a second option, not a second waveform : transform precoding is an extra stage in front of the same OFDM generator.
  • The reason is the power amplifier : DFT-s-OFDM lowers the peak to average ratio, which matters at a UE and does not matter at a gNB.
  • Numerology replaced waveform as the flexible part : one waveform with seven subcarrier spacings covers what a filtered waveform was proposed for.

Channel Coding at Tens of Gbps

Turbo codes carried LTE, and they scale badly. A turbo decoder is iterative and largely serial, so tens of Gbps means either a very large number of decoders running in parallel or a different code altogether. This question was settled early in the standardisation, and it was settled differently for data and for control.

Data went to LDPC. 38.212 sends every code block of the downlink shared channel to an LDPC encoder, and the uplink shared channel uses the same clause. An LDPC decoder is parallel by construction, because the parity checks are independent of each other. That is the property a turbo decoder does not have, and it is the whole reason for the change.

Two base graphs were defined rather than one. The choice between them depends on the payload size and on the code rate. Small payloads and low code rates go to base graph 2, and everything else goes to base graph 1. One graph tuned for a 100 kbit transport block wastes a great deal on a 100 bit one, so the specification carries both.

Control went to Polar coding instead. Downlink control information is encoded that way, and the PBCH payload is encoded the same way. Control blocks are short and need a very low error rate, and short blocks are where Polar codes perform well and where LDPC does not.

A third scheme covers the shortest payloads of all. 38.212 has a clause for small block lengths, with separate rules for one bit and for two bits. A HARQ acknowledgement is a single bit, and no block code is worth applying to it.

So NR carries three coding schemes where LTE carried turbo plus convolutional. The selection is made by payload length and purpose rather than by channel. So the same PUCCH can be encoded three different ways, depending on what it carries.

  • LDPC for data, Polar for control : the split follows block length, because the two codes are strong at opposite ends of it.
  • Parallelism was the deciding property : LDPC parity checks are independent, and that is what a turbo decoder cannot offer at these rates.
  • Two base graphs, chosen by size and rate : base graph 2 for small payloads and low code rates, base graph 1 otherwise.
  • A third scheme for one and two bits : the smallest control payloads use the small block length rules rather than either main code.

Initial Access with Narrow Beams

The path loss described in the millimetre wave section has a consequence that extends beyond the data channels. A cell that needs a narrow beam to reach a UE cannot broadcast its synchronisation signal in every direction at once. Every earlier generation assumed it could, and initial access was built on that assumption.

In LTE the synchronisation signals go out once per period across the whole cell. A UE searches for them knowing nothing about where it is, and one transmission serves every direction. Narrow the beam and that stops working, because a UE outside the beam hears nothing at all.

NR sweeps rather than broadcasts. The SS/PBCH blocks of one burst are sent in different directions one after another, and a UE finds the cell on whichever of them reaches it. Which of the candidate positions a cell actually uses is signalled by ssb-PositionsInBurst.

The size of that field is the beam budget, and 38.331 offers three shapes for it. They are shortBitmap at 4 bits, mediumBitmap at 8 bits, and longBitmap at 64 bits. The 64 bit form belongs to the higher frequency range, which is where the beams are narrow and where many of them are needed.

The cost is time. A cell sweeping 64 directions spends 64 occasions to cover what one transmission covered before, and a UE arriving at the wrong moment waits for the right one. That delay is the cost of the link budget, and it applies to every initial access.

The UE also has to report which beam it heard, or the gNB would have to sweep its response as well. That is what the SSB to occasion mapping in the RACH section above does. The preamble a UE sends identifies the beam it chose, and the answer comes back on that beam alone.

  • Direction became part of initial access : the mmWave link budget was answered by sweeping the synchronisation signal, not by raising power.
  • The bitmap is the beam budget : 4, 8 or 64 candidate positions, with the 64 bit form reserved for the high frequency range.
  • Sweeping costs time on every access : a UE may wait most of a burst before the beam covering it is transmitted.
  • The preamble carries the beam back : mapping RACH occasions to SSB indices is what lets the response be sent narrow as well.

Battery Life at These Data Rates

Every section above adds work for the UE. More bandwidth to sample, more antennas to process, shorter slots to decode, and more control channel to watch. All of it draws current, and the device carries the same battery it carried before. This is the constraint that shapes more of NR than any single feature does.

The largest cost is not receiving data. It is looking for control. A UE monitoring PDCCH in every slot at 120 kHz subcarrier spacing wakes eight times per millisecond, and it does so whether or not anything is addressed to it. Most of that work finds nothing.

Discontinuous reception is the first answer, and it is inherited from LTE rather than invented for NR. The field drx-onDurationTimer sets how long the UE watches at the start of a cycle, and drx-InactivityTimer extends the awake period after it is actually scheduled. Separate retransmission timers exist for the downlink and the uplink, so a pending retransmission keeps the receiver awake without keeping everything else awake.

Release 16 added a signal that arrives before the cycle begins. DCI format 2_6 notifies power saving information outside DRX Active Time, and its CRC is scrambled by PS-RNTI so that one transmission reaches a group of UEs. A UE told that nothing is coming skips its on duration entirely, and never wakes at all.

The bandwidth part is the second answer, and the section on dynamically configurable spectrum has already described it. An active bandwidth part narrower than the carrier means fewer samples per second through the converter. That is the ADC section above, seen from the battery rather than from the sampling theorem.

One pattern is common to all of it. NR did not lower any of the peak requirements this page lists. It arranged for the peak to be rare, and for the UE to spend most of its time doing something much easier.

  • Control monitoring costs more than data : a UE checks PDCCH on a schedule, and most checks find nothing addressed to it.
  • DRX sets when to look at all : an on duration timer, an inactivity timer, and separate retransmission timers per direction.
  • DCI format 2_6 removes the on duration : a group addressed signal before the cycle lets a UE skip waking up completely.
  • A narrow bandwidth part is a battery feature : fewer samples per second is the same saving the converter section describes, read from the other end.
  • The peaks were kept and made rare : none of the requirements above were relaxed, and the saving comes from how seldom they are met.

Reference

The paragraphs added under each heading above are checked against the published specifications rather than against the state of discussion at the time this note was written. The four documents below are the ones they quote.

[TS1] 3GPP TS 38.211 V19.4.0 - NR; Physical channels and modulation. Numerologies, slots per subframe, pi/2-BPSK, CSI-RS ports and the 64 PRACH preambles

[TS2] 3GPP TS 38.214 - NR; Physical layer procedures for data. The 1024 QAM MCS table, the Type I codebooks, and the mapping type B symbol allocations

[TS3] 3GPP TS 38.331 V19.3.0 - NR; RRC protocol specification. The RACH configuration fields named above

[TS4] 3GPP TS 38.101-2 - NR; UE radio transmission and reception, Range 2. The frequency ranges, the channel bandwidths and the EVM requirements

[TS5] 3GPP TS 38.212 V19.4.0 - NR; Multiplexing and channel coding. The LDPC and Polar clauses, the base graph selection, and DCI format 2_6

[TS6] 3GPP TS 38.213 V19.4.0 - NR; Physical layer procedures for control. The candidate SS/PBCH block positions used by the sweep