LATENCY — HOW DOES IT IMPROVE WITH 5G
Note: I have researched and assembled some of the content through multiple sources including 3gpp, and adapted it to make it simple to understand for all of us. If you are an enterprise and reading this, it should give you a view into the latency improvements on 5G and some granular details on how we arrive at it. This should further help in your research.
There is a lot of talk about 5G bringing in the capability of low latency transmission. A lot is said in marketing presentations and other technical literature. During my discussions with the customers, I often get asked about latency, and most of the time I breeze past it quoting the obvious “ 5G offers very low latency “. This is a humble attempt to actually document how this is achieved. There is a lot more to this subject, but this article will be kept as simple as possible so that people who are really curious can get this concept over a coffee. Note: some knowledge of the radio domain maybe required.
LTE and 5G technologies use OFDMA “Orthogonal Frequency Division Multiple Access” as an access scheme. It’s a multiplexing method, where multiple users can simultaneously send data. OFDMA takes an RF (Radio frequency) spectrum (ex:20 Mhz) and divides it into multiple sub-carriers of certain bandwidth closely spaced, a technique known as subcarrier spacing. Each subcarrier when modulated will carry that data signal, orthogonally separated from each other. This means when one signal reaches its peak the adjacent data signals are at their low zero points, therefore called orthogonally separate. The orthogonality of this scheme allows each subcarrier to operate without any interference from the other.
The scheme allows assigning multiple users a set of radio resources (frequencies in time) to conduct simultaneous data transmissions. Radio resources can be allocated to a userbase separately for uplink and downlink, using resource blocks called PRB (Physical Resource Blocks), where each resource block is a set of subcarriers grouped together.
Physical Resource Blocks (PRBs)
As stated above, resource blocks are blocks of subcarriers grouped together. Each subcarrier in the resource block carries data (technically called Symbols — basically modulated signal, sampled at a sampling rate — ex: 64 QAM — Quadrature Amplitude Modulation). The simplest way to understand this, QAM takes a signal (sinusoid), and samples it for digital values 1’s & 0’s -bits.
As an example, In LTE, a 20 Mhz RF block is divided into subcarriers, each subcarrier is 15 Khz in bandwidth. The picture below shows the representation of radio resources — in the frequency and time domain and the resulting resource blocks for LTE.
LTE Physical resource block (PRB) as shown above, is composed of 12 subcarriers.
12 Subcarriers x 15 Khz/Subcarrier = 180 Khz block of bandwidth available to carry data at a certain time duration, in this case, 1 millisecond.
Now, how many PRBs can we have?
This really depends upon the amount of usable spectrum available. This is allocated to a wireless carrier as part of their spectrum allocation from FCC.
Suppose you have 20 Mhz available. The picture below shows how that band is broken down.
Guard bands are introduced between two blocks of spectrum to avoid interference, but not between the subcarriers (Orthogonality).
So the first step is to calculate the Guard band, to arrive at the actual “usable” spectrum.
After accounting for the guard bands on both sides, you will be left with X Mhz bandwidth and a maximum of XYZ resource blocks. The calculations are below.
The following formula is used for calculating the Guard band
Guard band = (Available bandwidth — (N * PRB -PRB )) / 2
N= Bandwidth/PRB = 20000 Khz /180 Khz = 111.11 khz= 111 khz
Guard band = (20000 — (111.(180) — 180)) /2 = 200/2 = 100 khz
Total Bandwidth available after guard bands on both sides = 20,000 -200 = 19800 khz
Number of resource blocks will be = 19800khz/180 khz = 110 PRBs **
** There is some ongoing online discussion. Some claim this value to be 100, instead of 110.
Number of subcarriers in a 20 Mhz spectrum = 110 *12 = 1320
LTE Technology supports a maximum spectrum of 20 Mhz, however, using carrier aggregation features can go up to 100 Mhz.
So how different is 5G?
But 5G NR introduces something more interesting. 5G brings in the concept of flexible subcarrier spacing. In addition to 15 khz, 4 additional subcarrier spacing 30khz, 60khz, 120 kHz, and 240khz are introduced.
This subcarrier spacing flexibility is available to be utilized as below.
- 15 Khz, 30 Khz, 60 Khz is used in the FR1 frequency spectrum — that is under 6 Ghz.
- 120 and 240 Khz is used with FR2 frequency spectrum — 24 Ghz and above.
When it comes to 5G NR (New Radio) we get to play with channel bandwidth from blocks of 20 Mhz up to 400 Mhz.
Now, let’s calculate the number of PRBs for 5G with 30 Khz Subcarrier spacing.
The calculation for the number of PRBs for 20 Mhz and 40 Mhz available channel bandwidth is below:
For 20 MHz bandwidth with carrier spacing of 30Khz
Guard band = (bandwidth — ( N . df -df))/2 )
N= Bandwidth/PRB = 20000 Khz /360 Khz = khz= 55 khz
Guard band = (20000 — (55.(360) — 360))/2 = 560khz/2 =280khz
Total Bandwidth available after guard bands on both sides = 20,000- 560= 19440 khz
The number of resource blocks will be = 19440 kHz/360 kHz = 54 PRBs
Number of subcarriers = 54 *12 = 648
For 40 MHz bandwidth with carrier spacing of 30Khz
Guard band = (bandwidth — ( N . df -df))/2 )
N= Bandwidth/PRB = 40000 Khz /360 Khz = 111 khz
Guard band = (40000 — (111.(360) — 360))/2 = 400 khz/2 =200
Total Bandwidth available after guard bands on both sides = 40,000- 400= 39600 khz
The number of resource blocks will be = 39600 khz/360 khz = 110 Resource blocks
Number of subcarriers = 110*12 = 1320
How much Data can the PRBs carry?
OFDM symbol represents the data that is transmitted or received per subcarrier (ex: subcarrier 1) per time period (0.5 milliseconds). In the diagram below (marked in Red), 7 Symbols are transmitted in each of 71.4 microseconds in duration on subcarrier 1 slot 1. Each Symbol is divided into two parts. A Cyclic Prefix(CP) + Data.
The Data part takes 66.67 microseconds (1/15khz), whereas the rest of the time is occupied by the CP.
In this case 14 Symbols per 1millisec in the time domain, with a subcarrier spacing of 15khz.
(Why 14 → 1/15 kHz = 66 microseconds. Each symbol is 66 microseconds long + time for the cyclic prefix. That is approximate, 15 symbols per second without CP time. Minus some additional overhead, only 14 symbols per carrier in a 1-millisecond time slot can be accommodated)
Each symbol is sampled using 64 QAM (Quadrature Amplitude Modulation), which results in 6 bits of information (Assuming this for our calculation, but in reality, this really depends upon the Modulation and the coding scheme (MCS) discussed later).
Each 1 millisecond of the time slot, carries 6 bits x 14 Symbols = 84 bits of information.
Each millisecond slot carries →14 symbols,
Per second → 14 x 1000 = 140000 symbols/second,
Using 64 QAM, if we derive 6 bits per symbol — -> 14000 x 6 = 84000 bits/sec,
With 12 subcarriers per PRB → 84000 x 12 = 1008,000 bits/sec,
For 20 Mhz bandwidth we calculated 110 PRB — -> 1008,000 x 110 =110,880,000 = 110 Mbits/Second.
So the total throughput for a 15 khz subcarrier spacing, for a 20 Mhz channel bandwidth — 110 Mbits/second
What happens in 5G?
As we know 5G allows flexible subcarrier spacing or flexible numerology. Governed by the notation “μ “
In a 10msecond 5G frame ( same as in LTE), the μ determines the number of slots that will be made available in the one-millisecond subframe.
2𝝻 = 0 = 1 slot = 1 full millisecond
2𝝻 = 1 = 2 slot
2𝝻 = 2 = 4 slot
2𝝻 = 3 = 8 slot
2𝝻 = 4 = 16 slot
Depending upon “𝝻”, each slot now can carry 14 Symbols each for that time duration.
In the same 20 Mhz bandwidth example,
if the numerology used were “𝝁” = 1, we will have 2 timeslots in a 1 msec interval. Each slot carrying 14 symbols — now for 0.5 milliseconds.
This is with a subcarrier spacing of 30khz. This is shown below.
So now, 1 millisecond, carries 6 bits x 28 Symbols (14 x 2 slots)= 168 bits of information.
Compare this with what we got for LTE with the same bandwidth, but lower subcarrier spacing (For 15Khz where, “𝝁” = 0, giving us only 1 slot — we can send only 14 symbols in on millisecond).
In 5G, when“𝝁” = 1, with 30Khz we can send 28 symbols, doubling the symbol rate in 1 millisecond. We can now transmit twice the amount of data in one millisecond compared to what we saw in LTE using this flexible numerology technique.
Therein lies the benefit for end-user applications. With higher numerology you get a higher symbol rate at a millisecond-level — -> therefore information can be sent faster. The end Applications can receive that information faster, process it faster over the radio channel.
The same logic continues when “𝝁” = 2, 3, or 4, offering lower latencies for data transmission. For example, when “𝝁” = 2, the millisecond slot now accommodates 4 slots — each slot carrying 14 symbols. That is 4 x 14 = 56 symbols in ONE millisecond. A drastic improvement in latency.
Let’s look at 5G overall throughput,
Each millisecond slot carries →28 symbols, assuming “𝝁” = 1,
Per second → 28 x 1000 = 28000 symbols/second,
Using 64 QAM, if we derive 6 bits per symbol — -> 28000 x 6 = 168000 bits/sec, OR 28 x 6 = 168 bits in one millisecond, or one slot in a 10 millisecond frame. That is a drastic improvement over LTE which can only accommodate 14 x 6 = 84 bits in one millisecond.
With 12 subcarriers per PRB → 168000 x 12 = 2016000 bits/sec,
For 20 Mhz bandwidth we calculated 54 PRB — -> 1008,000 x 54=54,432,000 bits/sec = 54 Mbits/sec.
So for 5G the total throughput for a 30 khz subcarrier spacing, for a 20 Mhz channel bandwidth — 54 Mbits/second
With LTE we were limited to 15Khz as the subcarrier spacing, therefore just one slot within a millisecond timeslot will allow us to send only 14 symbols per millisecond. With 5G, with flexible subcarrier spacing, we can increase the number of symbols in a slot by increasing the subcarrier spacing, for example (doubling the amount of data in one millisecond with ex: 30khz spacing) improving latency. The process continues as we go higher on the subcarrier spacing (up to 240 kHz in case of millimeter-wave) and we will be able to pack in more symbols into a millisecond timeslot, thereby lowering the latency even further.
Enterprise applications now have this capability available. We need to make 5G implicit, integrated with the enterprise application basic design construct. One way of doing that is by creating integrated solution maps, where the solution is “tuned” right from the device, the network, and the edge/public application. That will help preserve the integrity of the entire use case, and help deliver data securely, accurately, and process for actions on a realtime basis. Hope this helps.
In my next article, I will take an actual use case and map it to this discussion. Standby.
Note on Throughput
In the case of LTE, we saw 110 Mbps and 5G at 54 Mbps. That seems too low because I have assumed 6 bits with 64 QAM theoretically. However in reality wireless systems employ much better modulation and coding schemes (MCS), and there is a published index that tells us a theoretical value of the number of bits that can be transmitted depending upon various conditions. This is a whole another subject.
The Index is here — → http://anisimoff.org/eng/lte_mcs.html. If we substitute those values with the right CQI (Channel Quality Indicator) + The antenna module — MIMO (2 x2 or 4 x4), we can approach higher theoretical speeds.
Note, not all of this throughput is available for User plane traffic, part of it is used by the signaling, and also depends upon user density at that location, the CQI values based on coverage at that specific location. Hopefully, I will be able to cover that in my next article.