230 when a majority of available resources for the time slot are allocated to the wireless device 102. In some embodiments, the time slot classification module 206 calculates a maximum transport block size (TBS) value for the time slot based on the communication channel characteristics and compares a TBS value of a grant allocation for the time slot to the wireless device 102 to the maximum TBS value to determine whether to characterize the time slot as an active time slot 230. In some embodiments, the time slot classification module 206 calculates a ratio of i) a TBS value of a grant allocation to the wireless device 102 for the time slot to ii) a calculated maximum TBS value for the time slot and compares the calculated ratio to a ratio threshold value to determine whether to classify the time slot as an active time slot 230. In some embodiments, the time slot classification module 206 classifies the time slot as an active time slot 230 when the calculated ratio satisfies, e.g., exceeds, the ratio threshold value. In some embodiments, the ratio threshold value is at least one-half and less than one.

The time slot classification module 206 characterizes a time slot as a last active time slot 234, when the time slot has a non-zero grant allocation that does not satisfy the threshold value. In some embodiments, the time slot classification module 206 characterizes the time slot as a last active time slot 234 when less than a majority of available resources for the time slot are allocated to the wireless device 102. In some embodiments, the time slot classification module 206 characterizes the time slot as a last active time slot when the calculated ratio of i) the TBS value of a grant allocation to the wireless device 102 for the time slot to ii) the calculated maximum TBS value (TBS_max) for the time slot does not satisfy, e.g., does not exceed, the ratio threshold value. For example, with a ratio threshold value T, the time slot classification module 206 characterizes a time slot as an active time slot 230 when TBS/TBS_max>T and as a last active time slot 234 when TBS/TBS_max≤T. TBS_maxcan be calculated based on channel quality characteristics (e.g., inferred from CQI values and corresponding MCS values, maximum RF bandwidth, as well as additional device capabilities, such as number of MIMO layers available, etc.). The ratio threshold value T can be configured as a non-zero value in the interval (0,1].

The time slot classification module 206 characterizes a time slot as a busy time slot 232, when the time slot has a zero grant allocation while there is downlink data traffic or uplink data traffic queued for the wireless device 102. As the wireless device 102 does not have access to downlink queue information for the base station 104, the wireless device 102 infers that a time slot with a zero grant allocation that occurs within a traffic burst that includes at least one active time slot 230 and a last active time slot 234 to be a busy time slot 232, where the base station 104 grants available resources for the time slot to other wireless devices 102.

The time slot classification module 206 characterizes a time slot as an idle time slot 236, when the time slot has a zero grant allocation and there is no downlink data traffic or uplink data traffic queued for the wireless device 102. The wireless device 102 infers that a time slot with a zero grant allocation that occurs between traffic bursts to be an idle time slot 236.

The time slot classification module 206 provides the classified time slots 208 to a latency inference module 210 which determines individual latency estimates 212 for time periods of busy time slots 232 within traffic bursts. A traffic burst includes at least a first active time slot 230 and a last active time slot 234. A traffic burst can include multiple active time slots 230. A traffic burst can include one or more busy time slots 232, which can be used to estimate latency for backlogged data traffic that is not able to be communicated to or from the wireless device 102 during the busy time slots 232. Traffic bursts can occur back-to-back or can be separated by one or more intervening idle time slots 236. The wireless device 102 does not attribute latency to time periods of idle time slots 236 between traffic bursts. The wireless device infers latency from time periods of busy time slots 232 that occur within traffic bursts. Latency can be determined as a time gap (time duration) between two active time slots 230 or between an active time slot 230 and a last active time slot 234 within a traffic burst and reflects a waiting time for the wireless device 102 to obtain a grant allocation for transmission or reception of data traffic. The latency inference module 210 can measure time periods of one or more busy time slots 232 within traffic bursts to generate latency estimates 212 that are provided to a latency monitoring module 214. In some embodiments, the latency estimates 212 output by the latency inference module 210 are time durations for each time period of one or more consecutive busy time slots 232 within traffic bursts. In some embodiments, the latency inference module 210 averages together individual latency estimates for individual time periods of one or more consecutive busy time slots 232 within a traffic burst and outputs a single averaged latency estimate 212 for each traffic burst.

The latency monitoring module 214 processes the individual latency estimates 212 to generate a set of one or more latency statistics 216. For example, the latency statistics 216 can include filtered latency estimates such as a moving average value for the latency estimates 212 and/or a latency probability distribution. The latency monitoring module 214 provides the latency statistics 216 to a latency mitigation module 218 that can trigger one or more latency mitigation actions 220 based on the latency statistics 216. Exemplary latency mitigation actions 220 include changing time slot aggregation parameters, configuring a regular (e.g., periodic) grant allocation for uplink transmission, changing settings (or preferences) for connected mode discontinuous reception (cDRX), adjusting settings (or preferences) for user equipment (UE) assistance information (UAI) release procedures, changing random access channel (RACH) procedures, and/or blocking select data traffic, e.g., background data traffic, to prioritize other data traffic, e.g., foreground data traffic. Additional exemplary mitigation actions 220 include switching RATs or prioritization for RAT use by the wireless device 102 and/or switching between cellular wireless and non-cellular wireless connections to improve data traffic throughput performance for the wireless device 102.

In some embodiments, the wireless device 102 sends one or more standardized UAI messages or proprietary messages to the base station 104 of the cellular wireless network to mitigate latency incurred by network congestion. A UAI message informs the base station 104 of various internal statuses of the wireless device 102 and allows the base station 104 to assign and/or control resources for the wireless device 102 based on the information indicated by the statuses in the UAI message. In some embodiments, a UAI message includes capability information for the wireless device, such as maximum settings, which can change dynamically over time. In some embodiments the wireless device 102 sends proprietary messages to the base station 104 to enable or disable various features to assist with improving latency for data packet communication with the wireless device 102.

FIG. 3 illustrates a diagram 300 of an example of a sequence of time slots over a traffic measurement time period 310 with the time slots identified with applicable classifications, e.g., by the time slot classification module 206. The traffic measurement time period 310 includes two traffic bursts 302 separated by several idle time slots 236. Each traffic burst 302 begins with an initial active time slot 230 and ends with a last active time slot 234. The first traffic burst 302 begins after an idle time slot 236 and includes the initial active time slot 230 followed by two additional active time slots 230 and then a latency time period 304 that includes two busy time slots 232 followed by an additional active time slot 230 and ending with the last active time slot 234. The second traffic burst includes two active time slots 230 followed by a last active time slot 234 with no busy time slots 232. The latency inference module 210 can determine and output a latency estimate 212 corresponding to the latency time period 304 for the first traffic burst 302. The latency inference module 210 can determine there are no latency estimates 212 to output (or in some cases, output a latency estimate with a value equal to a time duration of a single time slot) for the second traffic burst 302. The idle time slots 236 between traffic bursts 302 will not result in any latency estimates 212, as only traffic bursts are considered for determining latency estimates 212. In some cases, a traffic burst 302 can include multiple latency time periods 304 (not shown) and result in multiple latency estimates 212 corresponding to the multiple latency time periods 304 of the traffic burst 302. In some cases, the latency inference module 210 outputs multiple latency estimates 212 for a traffic burst that includes multiple latency time periods 304. In some cases, the latency inference module 210 outputs a single combined (e.g., averaged) latency estimate 212 for a traffic burst 302 that includes multiple latency time periods 304.

FIG. 4 illustrates a diagram 400 of an example of latency monitoring based on classification of time slots by a wireless device 102. A time slot classification module 206 classifies a sequence of time slots based on grant allocations to the wireless device 102 for the time slots and on a maximum grant allocation for the time slots based on communication channel characteristics for the cellular wireless access link 106 between the wireless device 102 and a base station 104. A latency inference module 210 identifies traffic bursts 302 based on the time slot classification, identifies latency time periods 304 within traffic bursts 302, and determines latency estimates 212 for the latency time periods 304. The latency estimates 212 can be averaged over a traffic burst 302 resulting an average latency estimate 402 for each traffic burst 302 that includes latency time periods 304. In some cases, a traffic burst 302 without a latency time period 304 does not result in a latency estimate 212 for the traffic burst 302. In some embodiments, a traffic burst 302 without a latency time period 304 results in a latency estimate 212 equal to a time duration of a single time slot for the traffic burst 302. The average latency estimates 402 for each traffic measurement time period 310 can be further averaged together to produce an average latency estimate 404 for each traffic measurement time period 310. The average latency estimates 404 for the traffic measurement time periods 310 can be processed to produce a long term weighted moving average latency estimate 406. The wireless device 102 can use the long term weighted moving average latency estimate 406 (or another statistical processing of the latency estimates 212) to determine whether to initiate one or more latency mitigation actions 220.

For the example shown in FIG. 4, a first traffic measurement time period 310-1 includes a first traffic burst 302-1 that includes two individual latency time periods 304, a first latency time period 304-1 that spans a single busy time slot 232 resulting in a latency estimate 212 with a latency value L₁, and a second latency time period 304-2 that spans two busy time slots 232 resulting in a latency estimate 212 with a latency value L₂. The two latency estimates 212 for the first traffic burst 302-1 can be combined to produce an average latency estimate 402 having an average latency value L_B1for the first traffic burst 302-1. The first traffic measurement time period 310-1 also includes a second traffic burst 302-2 that includes a single latency time period 304-3 that spans four busy time slots 232 resulting in a latency estimate 212 having a latency value L₃, which is also an average latency estimate 402 value L_B2for the traffic burst 302-2. The average latency estimates 402 for the first traffic measurement time period 310-1 can be averaged together to produce an average latency estimate 404 with a value of L_{AVE_1}for the first traffic measurement time period 310-1. The example of FIG. 4 further includes a second traffic measurement time period 310-2 having a first traffic burst 302-3 with no latency time periods 304 and a second traffic burst 302-4 with three latency time periods 302-4, 302-5, and 302-6 resulting in latency estimates 212 having values of L₄, L₅, and L₆, which can be averaged together to form the average latency estimate 402 with a value of L_B3for the second traffic burst 302-4. In some embodiments, the first traffic burst 302-3, with no latency time periods, results in a latency estimate 212 having a value equal to a time duration of a single time slot. The latency estimates 212 for the traffic measurement time period 310-2 can be averaged together to produce an average latency estimate 402 with a value L_{AVE_2}for the second traffic measurement time period 310-2. The average latency estimates 404 for each traffic measurement time period 310 can be combined using a weighted averaging formula with a weighting factor a to produce a weighted moving average latency estimate 406.

L_{WMA_T}=(1−α)L_{WMA_T-1}+αL_{AVE_T}

In an exemplary embodiment, the weighting factor is set to α=⅛. In some embodiments, each traffic measurement time period 310 spans two seconds. In some embodiments, the traffic measurement time period 310 is configurable by software/firmware in the wireless device 102. In some embodiments, a latency mitigation module 218 triggers one or more latency mitigation actions 220 when the weighted moving average latency estimate 406 exceeds a pre-determined latency threshold value. In some embodiments, the wireless device 102 maintains a history of the weighted moving average latency estimates 406 and estimates a latency probability distribution, e.g., a median value, a standard deviation, a maximum value, etc. Exemplary latency mitigation actions 220 include changing time slot aggregation parameters, configuring a regular (e.g., periodic) grant allocation for uplink transmission, changing settings (or preferences) for connected mode discontinuous reception (cDRX), adjusting settings (or preferences) for user assistance information (UAI) release procedures, changing random access channel (RACH) procedures, and/or blocking select data traffic, e.g., background data traffic, to prioritize other data traffic, e.g., foreground data traffic. Additional exemplary mitigation actions 220 include switching RATs or prioritization for RAT use by the wireless device 102 and/or switching between cellular wireless and non-cellular wireless connections to improve data traffic throughput performance for the wireless device 102.

In some embodiments, the wireless device 102 measures time gaps between grant allocations to estimate latency when a high application data rate occurs, which can imply an amount of backlogged traffic exists. Application data rates can be measured, in some cases, by monitoring TBS values. In some embodiments, the wireless device 102 measures time gaps between grant allocations to estimate latency when a high block error rate (BLER) occurs, which can imply multiple retransmissions are required and therefore there is backlogged traffic for the wireless device 102. In some embodiments, the wireless device 102 measures time gaps inside traffic bursts and uses TBS values and/or BLER values to confirm whether there is backlogged traffic and therefore latency can be inferred from the measured time gaps. In some embodiments, the wireless device 102 monitors a category associated with an application executing on the wireless device, such as gaming, augmented reality (AR), virtual reality (VR), voice over IP (VoIP), video streaming, audio streaming, media streaming, etc., and for certain application categories, the wireless device 102 measures time gaps between grant allocations to the wireless device 102 to estimate latency.

FIG. 5 illustrates a diagram 500 of an overview of another exemplary set of computational modules for an uplink latency tracking system in a wireless device 102. Latency monitoring in the downlink direction can use a traffic burst identification system at the wireless device as described herein to determine when there is backlogged downlink traffic queued at a base station 104 for transmission to the wireless device 102, because the wireless device 102 does not have access to queue status of the base station 104 for the data traffic for the wireless device 102. In some embodiments, the wireless device 102 supplements traffic burst identification with knowledge of uplink queues maintained at the wireless device 102 to monitor and estimate latency. An uplink traffic queue monitoring module 502 can provide an uplink queue status 504, which can be used to determine, at 506, whether there are queued (yet to be transmitted) uplink data packets and/or dropped data packets resulting in latency. In some embodiments, the wireless device 102 monitors uplink queues at the packet data convergence protocol (PDCP) layer. Uplink queue information 508 can be provided to an uplink latency determination module 510 that generates uplink latency estimates 512. An uplink latency monitoring module 514 can process the uplink latency estimates 512 to produce as set of uplink latency statistics 516, e.g., filtered uplink latency estimates, that can be provided to an uplink latency mitigation module 518 that uses the uplink latency statistics 516 to determine whether to perform (or recommend) one or more uplink mitigation actions 520. In some embodiments, the number of queued and/or dropped uplink data packets are monitored and compared to a configurable queue threshold value. When the number of queued and/or dropped uplink data packets satisfies (e.g., exceeds) the queue threshold value, an uplink latency estimate can be generated. In some embodiments, latency estimates are generated by monitoring time durations between uplink grants when the queue threshold value is satisfied. In some embodiments, latency estimates are generated by monitoring time durations between uplink grants when a block error rate (BLER) satisfies a BLER threshold value. In some embodiments, the wireless device 102 monitors buffer status report (BSR) requests to identify when there are backlogged data packets and generates latency estimates by monitoring time durations between uplink grants when a number of BSR requests satisfies (e.g., exceeds) a BSR threshold value.

FIG. 6 illustrates a flow chart 600 of an exemplary method for cellular wireless network latency estimation performed by a wireless device 102. At 602, the wireless device 102 determines occurrence of traffic bursts 302 on a cellular wireless access link 106 between the wireless device 102 and a base station 104, each traffic burst 302 including an initial active time slot 230 having a first non-zero grant allocation that satisfies a threshold value and a last active time slot 234 having a second non-zero grant allocation that does not satisfy the threshold value. At 604, the wireless device 102 estimates individual latency values 212 for time periods with zero grant allocation within the traffic bursts 302. At 606, the wireless device 102 processes the individual latency values 212 to obtain an estimated latency statistic 216. At 608, the wireless device 102 initiates at least one latency mitigation action 220 when the estimated latency statistic 216 satisfies a pre-determined latency threshold value.

In some embodiments, the wireless device 102 determines occurrence of traffic bursts 302 by classifying time slots having a non-zero grant allocation as an active time slot 230 when a majority of available resources for the time slot are allocated in the non-zero grant allocation to the wireless device 102, and as a last active time slot 234 when less than a majority of available resources for the time slot are allocated to the wireless device. In some embodiments, the method further includes the wireless device 102: i) calculating a maximum transport block size (TBS_max) value for each time slot based on communication channel characteristics; and ii) determining the majority of available resources for the time slot are allocated in the non-zero grant allocation to the wireless device 102 when a ratio of a transport block size (TBS) of the non-zero grant allocation to the TBS_maxvalue exceeds a ratio threshold value. In some embodiments, the ratio threshold value is at least one-half and less than one. In some embodiments, the wireless device 102 determines occurrence of the traffic bursts 302 by classifying a time slot having a zero grant allocation to be a busy time slot 232 when occurring within a traffic burst 302 and as an idle time slot 236 when occurring between traffic bursts 302. In some embodiments, individual latency values of a traffic burst 302 equal a time duration of one or more consecutive busy time slots 232 within the traffic burst