Low latency in 4.9G/5G

Transcription

1 Low latency in 4.9G/5G Solutions for millisecond latency White Paper The demand for mobile networks to deliver low latency is growing. Advanced services such as robotics control, autonomous cars and virtual reality will only become possible when ultra-low latency networks are widely available. There are a variety of techniques, existing and in standardization, that will reduce latency in both LTE and 5G networks. This white paper explores the components of latency and shows how they can be eliminated or their effects minimized.

2 Contents 1. Executive summary 3 2. Latency today and its evolution G latency solutions 7 Short TTI and mini-slot 7 Contention based access 8 Connectionless 9 Low latency during mobility G Latency solutions 10 Shorter TTI 10 Layer 2 latency reduction 10 Light connected Multi-access edge computing Conclusions Further reading Abbreviations 13 Page 2

3 1. Executive summary Low latency enables mobile networks to support new use cases. Future mobile networks will need to offer very low latency of the order of milliseconds. This Nokia paper illustrates the technologies that make low latency possible in 5G and LTE networks. Low latency is needed for industrial robot control, high frequency trading, power distribution network control, autonomous cars and virtual reality. An example use case demonstration is shown in Figure 1. A mobile network that supports low latency, will make these new services possible. The practical end-to-end latency in typical LTE networks is tens of milliseconds in the connected state and longer when starting from an idle state. New technologies are needed to bring latency down by a factor of ten to a hundred. Latency is improved by shorter transmission frame, by flexible resource allocation and contention based access, by connectionless protocols and by edge computing. These technologies will be included in 5G networks and can also be implemented in LTE networks. 5G will enable one millisecond latency in the radio. We also need to bring the content closer to the radio to achieve very low end-to-end latency using edge computing. LTE evolution can also reduce latency in the connected state to less than 2 ms. The importance of low latency in mobile networks has emerged during the last few years. When the first 3G HSDPA networks started in 2006, latency was more than 100 ms. Nokia showed in 2007 that HSPA evolution can bring latency below 25 ms. The general feedback at that time was clear - such low latency is impossible and would never be needed. Things are different now. The latest HSPA networks already provide latency below 20 ms, LTE offers nearly 10 ms and the need for even lower latency is clear. Figure 1. Nokia demonstration to illustrate the need for low latency for robot control. colors: Page 3 R 18 G 65 B 145 R0 G 201 B 255 R 104 G 113 B 122 R 168 G 187 B 192 R 216 G 217 B 218

4 2. Latency today and its evolution The aim in 5G radio is to provide a sub-1 millisecond round trip time, an ambition that is very challenging. 3G High Speed Packet Access (HSPA) networks can provide 20 ms latency in the best case, while current LTE networks can provide 10 ms. The improvement from 3G to 4G was two times while the target in 4.9G/5G is to improve latency by ten times from 4G. The main solution for minimizing the connected state latency is shorter Transmission Time Interval (TTI). HSPA TTI is 2 ms, LTE TTI 1 ms, 4.9G TTI 0.14 ms and 5G TTI ms. Shorter TTI makes the transmission time shorter but also shortens buffering and processing times. Shorter processing time also sets higher requirements for the receiver hardware and software. The latency components are shown in Table 1 and the latency evolution is illustrated in Figure 2. Best case measurements in a commercial network in Helsinki are shown in Figure 3: 19 ms in HSPA and 13 ms in LTE. Table 1. Round trip time components. HSPA LTE 4.9G 5G Downlink transmission 2 ms 1 ms 0.14 ms ms Uplink transmission 2 ms 1 ms 0.14 ms ms Frame alignment 2 ms 1 ms 0.14 ms ms Scheduling 1.3 ms ms 1 Pre-scheduled Contention based and pre-scheduled UE processing 8 ms 4 ms 0.50 ms ms BTS processing 3 ms 2 ms 0.50 ms ms Transport + core 2 ms (including RNC) 1 ms 0.1 ms (local content) 0.1 ms (local content) Total 20 ms ms 1.5 ms 1.0 ms 1 Scheduling period + capacity request + scheduling decision + PDCCH signaling. 2 Just Shared Control Channel (SCCH) End-to-end latency Transport + core BTS processing UE processing Scheduling Buffering Uplink transmission Downlink transmission 0 HSPA LTE 4.9G 5G Figure 2. Round trip time evolution from 3G to 5G. LTE latency ms ms HSPA latency ms ms Figure 3. Example speed test measurements in Telia network in Finland. Page 4

5 Figure 4 shows round trip time latency measurements by OpenSignal in the USA. The average latency is 60 ms, indicating that the backhaul and internet are the main sources of latency. There is no real benefit from 5G radio in this configuration. For low latency, local content or local breakout will be needed. 100 Latency in US Networks (OpenSignal 4Q/2016) ms Operator 1 Operator 2 Operator 3 Operator 4 Figure 4. Round trip time latency measurements in OpenSignal data. There are further delay components that need to be addressed - resource allocation latency and Radio Resource Control (RRC)/enhanced Radio Access Bearer (erab) setup. When RRC connection is available but no uplink resources are allocated, the user equipment (UE) must send a capacity request to the base station to obtain a capacity allocation. The delay components are shown in Table 2. The additional latency caused by the resource allocation is 18 ms in the case of a scheduling request period of 20 ms. The delay varies between 8 and 28 ms. The total delay for the packet is then 18 ms plus the round trip time. 3GPP allows a shorter scheduling request period (1, 2, 5 and 10 ms) but causes PUCCH capacity issues. Waiting for scheduling period (0..20 ms) Scheduling request transmission in uplink enodeb scheduling PDCCH transmission PUCCH transmission wait Total additional latency Delay 10 ms (average) 1 ms 3 ms 1 ms 3 ms 18 ms Table 2. Resource allocation delay components for LTE. Page 5

6 If the LTE UE is in an idle state, there is an additional latency caused by the establishment of RRC connection and setup of the erab. Therefore, the total delay (=setup + allocation + transmission) for the first packet is approximately 100 ms when starting from idle. The distribution of setup times in an LTE network is shown in Figure 5. The most typical value is 70 ms and most values are between 60 and 100 ms LTE setup time distribution (idle -> erab) Number of samples Milliseconds Figure 5. Setup time distribution from idle to erab including RRC setup. In short, the first packet transmission in LTE typically experiences a latency of ms and a lower latency of ms is achieved only when the uplink resources are already available. RRC idle Latency 100 ms because of RRC setup delay RRC connected but no uplink resources allocated Latency 30 ms because of scheduling delay RRC connected and uplink resources allocated Latency 10 ms Figure 6. Latency components in LTE radio today. Page 6

7 All these delay components are addressed in 5G and in 4.9G as shown in Figure 7. The connected state latency can be improved in 4.9G with shorter 2-symbol TTI, resource allocation with fast uplink access and setup time with faster signaling. 5G with a connected inactive solution and a contention based solution can provide even lower latency, including for the first packets. The detailed solutions are covered in the following sections. Connected with uplink resources Connected without uplink resources 4G 4.9G 5G Solution 10 ms 2 ms 1 ms Shorter TTI 30 ms <10 ms 1 ms Contention based access, fast uplink access Idle 100 ms <50 ms 1 ms Connected inactive state, shorter TTI, light connected Figure 7. Round trip time evolution from 4G to 5G. 3. 5G latency solutions Short TTI and mini-slot 5G supports various sub-carrier spacing and scheduling intervals depending on the bandwidth and on the latency requirements. Sub-carrier spacings of between 15 khz and 120 khz will be defined in Release 15. In later releases, sub-carrier spacing greater than 480 khz can be accommodated. With higher sub-carrier spacing, more symbols can be accommodated in a sub-frame, resulting in lower acquisition time. The narrow spacings are used with narrow 5G bandwidths and are better suited to extreme coverage due to the longer cyclic prefix. If we consider typical 5G deployment at the 3.5 GHz band, the bandwidth could be MHz, the subcarrier spacing 30 khz and minimum scheduling period 0.25 ms. The corresponding numbers in LTE are 20 MHz bandwidth, 15 khz subcarrier spacing and 1 ms scheduling period. 5G subcarrier spacing is designed to be 2^N multiples of 15 khz. 5G numerology is summarized in Table 3. Nokia has shown 1 millisecond latency using 5G AirScale and AirFrame products at 3.5 GHz. Sub-carrier spacing (khz) Spectrum <6 GHz <6 GHz <6..>20 >20 GHz Max bandwidth (MHz) Symbol duration (us) Nominal cyclic prefix (us) Scheduling interval (ms) Table 3. 5G flexible numerology. Page 7

8 The 5G specification includes the mini-slot concept, which allows for a very low latency. The minimum slot length in 5G is seven symbols and 14 symbols. Low latency with 15 khz sub-carrier spacing requires a shorter scheduling interval than seven symbols, a concept known as mini-slot. The frame structure is illustrated in Figure subframe (7 OFDM symbols) khz khz mmmm mini slot (2 OFDM symbols) 120 khz khz Figure 8. Frame structure for the mini-slot. Contention based access Contention based access refers to the uplink transmission where the UE autonomously sends data without any specific allocation or grant from the network. The benefit of this approach is minimized signaling, which improves latency and UE power consumption. The concept of contention based access is shown in Figure 9. The UE can be identified based on time and frequency resources and reference signal parameter. One option to implement contention based access is to apply Non-orthogonal Multiple Access (NOMA). The underlying idea is to allow different users to concurrently share the same physical resources, in either time, frequency or space. The NOMA concept is under study in 3GPP for 5G. The WCDMA uplink is essentially UE BS Data transmission without grant 1 st step = preamble + data Response 2 nd step = response Figure 9. Contention based access planned for 5G in 3GPP. Page 8

9 based on NOMA because it is non-orthogonal. The uplink transmission is simple in WCDMA when no time alignment or exact scheduling is required, but the uplink interference has proved to be a major issue in WCDMA in mass events with lots of uplink traffic. It is clear that the NOMA concept requires efficient uplink interference cancellation in the base station. Connectionless Connectionless solution in 5G refers to the case where the UE can maintain the RRC connection and core network connection at all times. Power consumption of the UE is minimized by introducing a new RRC connected inactive state. The latency will be very low because there is no need to create a neither an RRC connection nor erab. The current LTE networks typically release RRC connections after 5-10 seconds of inactivity to minimize UE power consumption. The connectionless solution is illustrated in Figure 10. LTE RRC idle RRC signaling RRC connected 5G Minimized signaling RRC idle RRC connected inactive RRC connected Figure 10. Connectionless 5G solution with RRC connected inactive state. Low latency during mobility Mobility has been the basic capability in mobile networks since 2G systems and was enhanced in 3G and 4G networks. 5G needs new mobility solutions to fulfill the new requirements for low latency and high reliability. The aim is to achieve 0 ms interruption time during mobility, requiring advanced make before break concepts. Another aim is to further improve robustness in multi-connectivity. Mobility in 5G not only refers to moving devices, but also static devices in changing environments. This applies in particular at mmw with passing cars, rotating bodies and obstructions, and also in industrial environments where many metal objects may be in motion. 5G includes solutions to support high speed trains up to 500 km/h even at 28 GHz frequency. The Doppler frequency is more than 10 khz and requires new solutions. Two specific solutions are Doppler compensation and decoupled control and data. A Doppler compensation algorithm estimates the Doppler shift caused by the rapid movement and corrects the received data symbols with estimated phase difference on a symbol-by-symbol basis in the time domain. Page 9

10 4. 4.9G Latency solutions Shorter TTI LTE evolution brings a number of solutions for minimizing latency in the connected state. Shorter TTI options are part of Release 15. The current TTI is 1 ms in LTE. Shorter 7-symbol = 0.5 ms TTI Shorter 2-symbol = 0.14 ms TTI Shorter time for UE processing and response defined in 3GPP 2-symbol TTI enables a round trip time of below 2 ms as shown in Figure 11. Nokia has demonstrated sub-two milliseconds latency using 4.9G technology with AirScale base station. Frame size Round trip time 1 ms 14 symbol TTI 10 ms Release 8 7 symbol TTI 5 ms Release 15 2 symbol TTI 0.14 ms 2 ms Release 15 Figure 11. Short TTI in LTE. Layer 2 latency reduction The resource allocation in LTE typically adds an 18 ms delay. It is possible to get around this delay by pre-allocating resources for the UEs in the uplink, although this feature is usually disabled. The problem is that the UE must send some data even dummy data if it gets an allocation in the uplink. Such dummy transmission increases uplink interference. Release 14 enhancements make it possible to give an uplink allocation to a UE that does not require it to send any data if it has nothing in the buffer. Therefore, it will be possible to pre-allocate uplink resources at least for a short period to minimize the latency. The allocation could be done, for example, always following the downlink transmission because an imminent response in the uplink is likely. This kind of layer 2 latency reduction concept is shown in Figure 12. Page 10

11 Pre-Release 14 Layer 2 latency reduction in Release 14 Wait for scheduling period Semi persistent scheduling Scheduling request Data Scheduling decision Grant on PDCCH Data Figure 12. Fast uplink access. Light connected Frequent switching between idle and connected modes increases signaling and latency. The light connected concept aims to improve this area. Release 13 uses the RRC resume concept where several signaling messages are still needed. Release 14 allows the RAN to maintain the connection to the EPC when the UE connection is suspended. Figure 13 shows the UE context retrieve procedure after UE mobility. The light connected concept would also need changes in the EPC. UE New enodeb Old enodeb MME RRC connection resume request UE context retrieve RRC connection resume request S1 path switch S1 release Figure 13. Context retrieve procedure in Light connected solution. Page 11

12 5. Multi-access edge computing Future network architecture is likely to include edge cloud techniques where the content can be cached for multi-access edge computing or where the local breakout can be provided to the local intranet or internet. The number of edge clouds needs to be more than the number of Base Station Controllers (BSC) or Radio Network Controllers (RNCs) today. The typical expectation is one local cloud supporting 100 sites or 100,000 subscribers. Low latency transport is preferred from the base station site to the local cloud. Antenna site Antenna RF Low layers Transport Ethernet transport Local cloud Higher radio layers Multiconnectivity Interference management Multi-access Edge Computing Distributed core Nokia AirFrame data center Nokia AirScale base station Figure 14. Network architecture with local cloud. 6. Conclusions The importance of low latency in mobile networks has grown in recent years. HSPA and LTE networks have achieved ever-lower latency performance, yet further gains will be vital as we move into the 5G era. 5G radio aims to provide a sub-1 millisecond round trip time, an ambition that is very challenging., requiring an order of magnitude improvement over 4G. Latency is improved by shorter transmission frame, by flexible resource allocation and contention based access, by connectionless protocols and by edge computing. These technologies will be included in 5G networks and can also be implemented in LTE networks. Page 12

13 7. Further reading Nokia white paper: 5G Master Plan Nokia white paper: 5G for Mission Critical Communication Nokia white paper: Translating 5G use cases into viable business cases Nokia white paper: Dynamic end-to-end network slicing for 5G white paper Nokia white paper: 5G System of Systems white paper 8. Abbreviations BSC BTS CoMP CQI EPC erab FDD HSPA LTE MIMO MU-MIMO NOMA PDCCH PUCCH RF RNC RRC SCCH TDD TM TTI TX UE Base Station Controller Base Station Coordinated Multipoint Channel Quality Indicator Evolved Packet Core Enhanced Radio Access Bearer Frequency Division Duplex High Speed Packet Access Long Term Evolution Multiple Input Multiple Output Multiuser MIMO Non-orthogonal Multiple Access Physical Downlink Control Channel Physical Uplink Control Channel Radio Frequency Radio Network Controller Radio Resource Control Shared Control Channel Time Division Duplex Transmission Mode Transmission Time Interval Transmitter User Equipment Page 13

14 Nokia is a registered trademark of Nokia Corporation. Other product and company names mentioned herein may be trademarks or trade names of their respective owners. Nokia Oyj Karaportti 3 FI Espoo Finland Tel (0) Product code SR EN Nokia 2017 Page 14