5G资源调度目标要求的提案

目前关于5G的资源调度这块已经有了相关的提案，涉及调度框架（Scheduling framework）和其它的相关需求（Related Requirements），本篇博文就对比分析一下目前5G的提案与LTE相比，在资源调度方面的相同点和不同点。

2016年5月，在江苏南京召开的3GPP例行会议期间，诺基亚、阿朗上海贝尔提交了一篇关于5G的调度框架和相关需求，文档编号R1-165023。下文表1摘录了其中的核心内容，蓝色字体是自己做的说明。

（表1）

2.1        Basic scheduler considerations The basic functionality of network controlled dynamic scheduling (as known also from LTE) is assumed to be adopted for 5G, where the basestation controls time-frequency transmission resources for the its users to transmit, or receive, transport blocks. The scheduling instructions from the basestation is issued to the UEs by transmitting downlink control information carrying UE specific scheduling information for downlink allocation and uplink transmission grants.Leading to the following proposal: ·  Proposal 1: Network controlled dynamic scheduling is the default scheduling mechanism, where the network informs the UE of the allocated time-frequency resources for transmission of transport blocks via downlink control information carrying UE specific scheduling information for downlink allocations and uplink grants. 阿米尔C：在协议规范的制定中，新的策略参考已有的、成熟的策略是很常见的方式。比如2G、3G、LTE不同制式之间小区选择或重选的策略大同小异。该提案中5G资源调度的基本策略，也将采用LTE的动态调度方式。LTE的动态调度方式就是，由网侧动态的发送不同种类的DCI来配置下行（如DCI1A、DCI2等）或上行（如DCI0）传输的资源授权。关于DCI的相关内容，请参考本空间的相关博文。 The scheduler shall be able to efficiently allocate radio resources to the users in coherence with their QoS requirements, available data for transmission, radio conditions, etc. This means that the scheduler shall support a large dynamic range in allocated radio transmission resources per grant, as well as transport block sizes. As an example, eMBB is known to have large dynamic range of the user plane data payload sizes to be scheduled,ranging from only several tens of bytes (e.g. for application layer control messages; TCP ACK’s, so-called “keep alive” message, etc.) to many mega-bytes of data for large data file transmissions. In fact, measurements from today’s LTE network with MBB traffic shows a significant fraction of small to medium size scheduled transport blocks. Similarly, mMTC and URLLC typically require scheduling of payloads from tens of bytes to thousands of bytes, depending on the exact application. Leading to the following: ·Proposal 2: The scheduler functionality shall support scheduling of payloads from few tens of bytes up to very high transport block sizes occupying the full carrier bandwidth using the highest modulation order and minimal error correcting coding. 阿米尔C：5G的不同业务也对应着不同的QoS，不同QoS的优先级不同，影响着调度的优先级。关于QoS的优先级，请参考博文《LTE-TDD资源调度（1）-QCI、GBR和Non-GBR》。与LTE的调度策略类似，该提案中5G的资源调度也需要考虑不同业务的QoS、无线链路的链路质量等这些参数。同时，当分配了全部的带宽资源、使用了最大的调制方式和误码率最小的时候，每次调度需要能够分配从几十字节到几十几百兆不等的数据资源。这里的最大吞吐量需求比LTE高了百倍左右的数量级，在LTE-R9里，每次动态调度最多也就1百多K的数据量。 As known from LTE, the scheduler shall support single-user and multi-user MIMO, i.e. scheduling of multiple transport blocks to the same or different users on the same (or partly overlapping) physical layer radio resources. For such cases with overlapping transmissions on the same time-frequency radio resources, options for assistance to support interference suppression and/or cancellation shall be investigated, leading to the following proposal: ·Proposal 3: Support for scheduling of overlapping transmissions on the same physical layer time-frequency resources to the same or different users shall be supported, including potential assistance to facilitate efficient interference suppression and/or cancellation.    阿米尔C：LTE支持相同的RB资源承载多个TB块，这样可以提升吞吐量。该提案中5G的资源调度，同样需要支持这样的功能。关于LTE的双流调度，请参考《LTE下行物理层传输机制（8）-DCI2A格式和下行双流的流量制约》等相关博文。 Efficient use of radio of radio resources is obviously a priority.This means the network should be able to schedule its users to efficiently utilize the available radio resources. Given Proposal 2, this sometimes means scheduling only a single user on the full system bandwidth (given there is sufficient data), to some-times scheduling larger number of users, some of which may require only few transmission resources for sending small to moderate size transport blocks. Hence, the scheduler shall support scheduling of only a single user up to simultaneous scheduling of a potentially large number of users. This essentially calls for flexible design of resources for transmitting scheduling grants, as the required transmission resources for scheduling grants will vary accordingly. In fact, it is a known problem from LTE that the system some-times suffers from control channel (PDCCH or E-PDCCH) blocking [2]-[5],preventing efficient use of all radio resources when desirable to scheduler larger number of users. For 5G, we should aim at a more flexible scheduling grant design that overcome such problems. The exact value for how many users should be simultaneously schedule per carrier naturally depends on the carrier bandwidth, carrier frequency, and anticipated service composition (detailed value is FFS). In summary, this leads to the following proposal: ·    Proposal 4: The scheduler shall support multiplexing of larger number of users (say N users) per carrier to ensure efficient radio resource usage. This calls for a highly scalable design of radio resources for transmitting scheduling grants to avoid control channel blocking.The exact value of N depends on the carrier bandwidth, carrier frequency, etc. 阿米尔C：在LTE里，每个调度周期需要支持多个用户的资源分配，但在实际的多用户并发业务场景中，很多时候并不是RB资源不够，而是CCE资源冲突导致无法分配DCI，从而导致无法为更多的用户分配资源。该提案中5G需要取消这种制约，每个调度周期的并发用户个数N，不应该再受限于控制信道。关于CCE的内容，请参考空间里的相关博文。 2.2        Potential of using different TTI sizes: It is well known from the existing literature that there are fundamental trade-offs between scheduling users to maximize their spectral efficiency, coverage, latency, or reliability [6]-[7]. In theory, this calls for flexible scheduler functionality that allows scheduling each link (user) incoherence with its desired optimization target. One option allowing the former,is to design 5G to support scheduling with different TTI sizes [8]. Scheduling latency critical data (e.g. URLLC) with a short TTI size obviously a priority. However, scheduling all users with a very short TTI is not optimal. Using long TTIs allows us to benefit from larger coding gains, and it also imposes lower control (scheduling grant) overhead. This comes, however, at the expense of latency increase; in that respect, the usage of longer TTIs is more beneficial for eMBB and mMTC users for which the required data rate may be high and the latency requirements are less stringent. Setting the TTI size per scheduling grant furthermore offers the possibility to optimize the eMBB services using the Transmission Control Protocol (TCP). During the initial data transmission session, the end-user experienced performance is primarily determined by the RTT due to the slow start TCP procedure (TCP flow control).Therefore, it would be advantageous to first perform scheduling of the eMBB TCP users with short TTIs, followed by longer TTI sizes when reaching steady state operation. On a similar note, it may be desirable to schedule a user with different TTI sizes in the uplink and downlink. As an example, a coverage limited UE in macro-cellular setting may require a TTI size of ~1ms to maintain uplink reception, while a shorter TTI size could be used in the downlink. Hence, the following proposal: ·  Proposal 5: The benefit of being able to schedule users with  variable TTI sizes shall be further assessed, including options for asymmetric link operation,where a user is scheduled with different TTI sizes in the downlink and uplink(especially relevant for macro-cellular scenarios at lower carrier frequencies).  阿米尔C：在LTE里，TTI周期固定是1ms。该提案提出，根据不同的业务需求，TTI周期将是可变的，比如可以减少到0.2ms或0.25ms。 2.3        UE scheduling grant monitoring: The scheduler framework for 5G shall naturally be designed to have reasonable UE complexity. That means that the search space for reception of scheduling grants shall be controllable, with reasonable UE complexity and energy(battery) consumption. In this context, it is proposed to consider efficient time-frequency domain DRX to avoid that UEs always have to search over the full system bandwidth and/or in all subframes for scheduling grants (see e.g. the example in [8]). As an example, it is unnecessary to have a UE with a moderate data rate connection monitor the full carrier bandwidth of e.g. 100Mz; while it would be sufficient to only monitor a fraction of the carrier bandwidth to save power. Similarly, a URLLC UE should monitor for scheduling grants on a fine time resolution to fulfil the corresponding latency requirements, while a mMTC and eMBB UE could be configured to monitor for scheduling grants on a coarser time-resolution. Finally, low cost mMTC devices may not have the capability to monitor large system bandwidths, and hence should be able to receive their scheduling grant on a fraction of carrier bandwidth only. These considerations leads to the following proposal: ·    Proposal 6: The UE effort for searching for scheduling grants (i.e. downlink control channel detection) shall be manageable. Both in terms of number of search options, energy consumption, etc.Shall support time-frequency domain DRX functionality to avoid that UEs always have to search over the full system bandwidth and/or in all subframes for scheduling grants.  阿米尔C：LTE只支持1.4M、3M、5M、10M、15M、20M这些带宽，但到了5G，则可能需要支持40M、100M这些超大带宽，如果还像LTE那样，每个UE都需要全频段盲检测PDCCH信道，显然是不合理的，既费电又增大了处理时间。所以提案提出，每个UE的PDCCH位置是可管理的，可以限制在整个带宽的某一个片段或区间里。 2.4        Additional considerations: In addition to independent per-cell scheduling of the corresponding users, the scheduler design shall be scalable to also support more advanced RAN implementations that e.g. utilize C-RAN with front haul connections and RRHs, or coordinated multi-cell scheduling over backhaul connections. Hence, the following proposal: ·      Proposal 7: The scheduler design shall offer the necessary flexibility to efficiently support multi-cell/multi-node coordinated scheduling, both for a distributed and centralized RAN implementation.  阿米尔C：提案指出5G的调度器，需要能同时支持分布式和集中式的多小区/多节点无线网络架构。 Finally, although our proposal is to assume dynamic scheduling as the default, the potential of semi-persistent scheduling (SPS) in 5G should not be discarded at this point in time. However, the driver for potential inclusionof SPS in 5G should probably not be VoIP as was the case for LTE. Instead, it is proposed to mainly consider the use of SPS for some of the MTC use case with modest data rates, where such techniques could offer benefits. This includes URLLC use cases, where SPS could offer benefits by reserving transmission resources for such latency critical traffic in advance. One particular advantage of SPS for URLLC is that scheduling instances are pre-allocated(configured with robust and highly reliable RRC signalling), and therefore does not rely on physical layer signalling of every scheduling allocation/grant. The former is of particular importance for fulfilling the ultra-reliability requirements[1] of 10e-5, where physical layer control channel errors would play a dominant role for traditional dynamic scheduling.    ·      Proposal 8: Inclusion of semi-persistent scheduling options is recommend. Is e.g. relevant for MTC use cases (including URLLC), rather than being optimized for VoIP as in LTE. 阿米尔C：在LTE里有可选择性支持的SPS功能，用于VoIP业务的周期调度。提案提出，5G同样需要可选择性支持的SPS功能，但此时SPS不是为了VoIP，而是为了支持MTC（Machine Type Communication）。  For basestation scheduling of a UE for uplink transmission, the scheduling grant includes a pointer to the radio resources that the UE shall use, as well as the modulation and coding, resulting in certain support transport block size (TBS). However, as the basestation may not always have accurate and up-to-date knowledge of the pending data in the UE for uplink transmission, the two following undesirable cases may happen:(i) the basestation schedule the UE with a too large TBS, or (ii) the basestation schedule the UE with a too small TBS. In LTE, the UE is mandated to use zero padding if the scheduled TBS is larger than the amount of pending data for uplink transmission in the device. This is inefficient from radio resource point of view, resulting in wasted capacity. The 5G new radio should therefore include a deterministic UE procedure for autonomous down-scaling of the used transmission resources to avoid zero padding, in cases where the scheduled TBS exceeds the amount of pending data in the UE. By including such a procedure,wasting of radio resources is avoided. The exact solution for UE autonomous down-scaling of the used uplink transmission resources naturally require further studies. Such solutions also enable the new 5G radio to better meet latency requirements of services such as URLLC. ·      Proposal 9: Options for UE autonomous down-scaling of the used uplink transmission resources in case of insufficient data to transmit as compared to the schedule transport block size shall be included. 阿米尔C：eNB是不知道UE具体的数据上传需求的，UE发出的BSR只是一个数值的范围，eNB分配的资源很容易超过UE实际使用的资源大小。在LTE里，这种情况会导致MAC PDU中出现大量的填充。到了5G，该提案提出为避免资源浪费，如果上行分配的资源超过UE待发数据需求，要避免出现填充。