How can latency be reduced in 5G compared to 4G? That's what we'll be looking at in this video. Let's start by looking at 4G. Transmission is organized over time in frames of 10 milliseconds or radio frames. A radio frame is itself broken down into 10 successive one-millisecond sub-frame. The sub-frames are filled with pairs of resource blocks. A pair of resource blocks is a set of 12 successive subcarriers in the frequency space and 2x7 successive symbols. To transmit a block of data, the minimum requirement is a pair of resource blocks. That means that data transmission takes at least one millisecond over a radio. In 5G, we still have the 10-millisecond radio frame and the one-millisecond sub-frame, whatever the numerology. But the definition of the resource block is more flexible. This is known as the Physical Resource Block or PRB. The 12 subcarriers are still there, but it's possible to transmit over a variable number n of successive symbols. This means that similarly to 4G, there can be up to 14 successive symbols per PRB, but that number can be lower. As the PRB is defined by a successive number of symbols, With low numerology, for example, here, numerology 0, 14 symbols will take one millisecond. With numerology 1, 14 symbols take half a millisecond. In the radio interface specification, the notion of timeslots is defined. A time slot corresponds to 14 successive symbols. The duration is, therefore, a function of the numerology, we get one milliseconds for numerology 0 to 1/16 milliseconds for numerology 4. The different numerologies can be mixed. Here, we can see an example. The band is divided in several parts, and only one numerology is used in every bandwidth part. Here, band numerologies 1 and 0 are used. Careful, mixing numerologies generates inter-numerology interference. There should be guard bands to limit this interference. We also need to look at FDD, which is Frequency Division Duplex. FDD use one carrier for downlink transmission and another frequency for the uplink, which means the transmission from the terminal to the eNB. With FDD, there is a static share of the spectrum. Just like in 4G, it's possible in 5G to use TDD or Time Division Duplex. The same frequency is used in uplink and downlink. In 5G, the share of the spectrum is highly flexible and dynamic for periods of 0-5 milliseconds, 0.625 of one millisecond, and so on. In this period, we can define duration, which will be reserved for downlink transmission, a duration reserved for uplink transmission, and a flexible duration, which can be used for downlink or uplink or no transmission at all. The particularity is that these durations are expressed as a number of symbols, which means that it's possible to have a transmission, for example, on the downlink for less than one slot or little more than five slots as we can see here. Secondly, if we consider the time slot for the highest numerology, the period can be very short. In numerology 3, we achieve 0.125-millisecond slots. This means that, in this case, there is one millisecond period in which we can make both downlink transmission and uplink transmission. For example, to transmit a block of data to a terminal and for this terminal to transmit its acknowledgment, all these will take less than one millisecond. In conclusion, in a 5G NR interface, time is divided into slots, sub-frames, and frames. The frame is always 10 milliseconds, a sub-frame is always one millisecond, and the slot varies in duration according to the numerology, bearing in mind that it's one over two to the power n milliseconds for numerology n. One millisecond for numerology 0, and 0.125 milliseconds for numerology 3. We still have the concept of Physical Resource Block, which includes a variable number of successive symbols, but always on 12 subcarriers. The duration of a PRB can therefore be less than one slot. Finally, there is the possibility of FDD and TDD. For TDD, allocation periods can be defined over relatively short periods. All of this mean that on the radio interface side, latencies of less than one millisecond are possible. Service latency includes all the latencies such are that caused by the transmission of packets over the core network. To conclude, here is a table recapping different numerologies and the principle values of the parameters. We will not be commenting on this table, but instead invite you to study it at your own pace.