Adaptive Transmission Parameters Selection Algorithm for URLLC Traffic in Uplink
Aleksei Shahsin, Andrey Belogaev, Artem Krasilov, Evgeny Khorov
AAdaptive Transmission Parameters SelectionAlgorithm for URLLC Traffic in Uplink
Aleksei Shahsin ∗† , Andrey Belogaev ∗ , Artem Krasilov ∗† and Evgeny Khorov ∗†∗ Institute for Information Transmission Problems, Russian Academy of Sciences, Moscow, Russia † Moscow Instutute of Physics and Technology, Moscow, Russiae-mail: { shashin, belogaev, krasilov, khorov } @wireless.iitp.ru ©2020 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, includingreprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, orreuse of any copyrighted component of this work in other works. Abstract —Ultra-Reliable Low-Latency Communications(URLLC) is a novel feature of 5G cellular systems. To satisfystrict URLLC requirements for uplink data transmission, thespecifications of 5G systems introduce the grant-free channelaccess method. According to this method, a User Equipment(UE) performs packet transmission without requesting channelresources from a base station (gNB). With the grant-freechannel access, the gNB configures the uplink transmissionparameters in a long-term time scale. Since the channel qualitycan significantly change in time and frequency domains, thegNB should select robust transmission parameters to satisfythe URLLC requirements. Many existing studies consider fixedrobust uplink transmission parameter selection that allowssatisfying the requirements even for UEs with poor channelconditions. However, the more robust transmission parametersare selected, the lower is the network capacity. In this paper,we propose an adaptive algorithm that selects the transmissionparameters depending on the channel quality based on thesignal-to-noise ratio statistics analysis at the gNB. Simulationresults obtained with NS-3 show that the algorithm allowsmeeting the URLLC latency and reliability requirements whilereducing the channel resource consumption more than twice incomparison with the fixed transmission parameters selection.
Index Terms —5G, URLLC, Grant-free, Uplink, MCS, K-repetition, parameters selection
I. I
NTRODUCTION
Ultra-Reliable Low-Latency Communications (URLLC) is anovel traffic type that will be supported by the next-generationcellular networks (5G) in addition to enhanced Mobile Broad-band (eMBB) and massive Machine-Type Communications(mMTC) [1]. Many applications, such as autonomous vehiclesinteraction or telesurgery, generate uplink URLLC-traffic [2].In particular, each autonomous vehicle transmits informationcollected from its sensors, e.g., information about its position,speed, acceleration, or obstacles detected on the road. Sim-ilarly, during the telesurgery, a surgeon transmits commandsto remote robotic manipulators that react to these commandswith corresponding actions. To ensure the efficient work ofsuch applications, they impose very strict latency (severalmilliseconds) and reliability (higher than . ) require-ments [3]. It should be noted that problem of quality of serviceprovisioning for these applications is not only considered forcellular networks, but also attracts much attention from the The research has been carried out at IITP RAS and supported by the grantNo 18-37-20077 mol-a-ved of the Russian Foundation for Basic Research. wired and Wi-Fi communities [4]. In this paper, we focus onsolution for 5G cellular networks.The 3rd Generation Partnership Project (3GPP) specifica-tions, e.g., [5], for 5G networks describes two methods for anuplink channel access: grant-based and grant-free (configuredgrant). According to the grand-based method, to access anuplink channel, a user equipment (UE) transmits a schedulingrequest and waits for a scheduling grant from a base station(called gNB). When the grant is received, the UE starts trans-mitting data. As a result, the grant-based channel access delaysthe actual transmission, which makes this method inapplicableto most of URLLC use cases. Hence, the grant-free channelaccess is usually considered for URLLC data transmission.With the grant-free channel access, a gNB selects thechannel resources and the transmission parameters, e.g., Mod-ulation and Coding Scheme (MCS), for each UE in a long-termtime scale. To improve reliability, UEs can perform multipletransmission attempts. Before the transmission attempt, a UEcan either wait for the feedback from a gNB related tothe previous attempts or perform the transmission withoutwaiting for feedback. In this paper, we consider the schemewithout feedback called K-repetition, which implies that aUE makes K transmission attempts for each data packetusing the parameters configured by the gNB. Since the K-repetition does not require receiving feedback from the gNBafter each transmission attempt, it allows significantly reducingthe data transmission latency. To improve the reliability, thegNB uses Hybrid Automatic Repeat reQuest (HARQ) scheme,e.g., Chase Combining (CC) [6]. Many works compare K-repetition with other schemes that use the feedback from agNB using analytical models [7], [8] and simulations [9], [10].Their results show that K-repetition is more effective in termsof transmission latency and reliability except for the casesof high load with massive overlapping of transmissions, i.e.,when transmissions corresponding to different UEs with highprobability use the same channel resources.Since the channel quality can significantly change in timeand frequency domains, and the gNB selects the transmissionparameters for relatively long periods of time, a UE shoulduse robust transmission parameters to satisfy strict URLLCquality of service requirements. In this case, the minimizationof channel resource consumption is challenging because themore robust parameters are used, the more channels resourcesare consumed, which leads to lower network capacity, i.e., a r X i v : . [ c s . N I] F e b he number of data flows that can be served simultaneously.Studies [7]–[10] consider usage of fixed MCS and the numberof transmission attempts, so they do not consider the dynamicparameter selection problem. The papers [11], [12] studiesthe selection of K to maximize the number of users inthe network. They consider that in the case when the sameresources are assigned to multiple UEs, simultaneous trans-missions of different UEs may lead to unsuccessful transmis-sions. However, in both works, the probability of unsuccessfultransmission is estimated without taking into account that theprobability of unsuccessful transmission depends on the MCS.The channel resource allocation problem for K-repetition isconsidered in [13]. The authors propose to divide availablechannel resources into multiple groups and randomly select agroup for each transmission attempt. They propose to selectMCS corresponding to the number of these groups (highernumber of groups corresponds to higher code rates). This studyshows that the K-repetitions scheme can provide the requiredpacket loss probability with the considered in the papertransmission parameters values. However, the authors do notprovide any adaptive method for selecting these parameters,e.g., depending on the channel quality. The paper [14] providesthe algorithm for selecting K and the channel resources fortransmission. However, this work considers fixed probabilitiesof successful decoding without taking into account that theydepend on the used MCS. The paper [15] provides the numberof transmission attempts (i.e., K ) selection algorithm basedon an estimation of fading correlation function. The authorssuggest using two attempts in the case of low correlation andfour in the case of high correlation. Numerical results showthat this method can increase reliability and resource efficiencyin a multi-user scenario compared to using a fixed number ofattempts. However, this work does not consider the problemof selecting MCS and proposes to use a fixed robust MCS.The provided above literature analysis show that the existingpapers do not provide any adaptive method for selecting MCSdepending on channel conditions. In this paper, we proposethe adaptive transmission parameters (i.e., MCS and number oftransmission attempts) selection algorithm based on estimatingof the packet loss ratio for each parameter configuration usingthe Signal-to-Noise Ratio (SNR) statistics at the gNB.The rest of the paper is organized as follows. In Section II,we introduce the proposed algorithm. In Section III, we evalu-ate its preference with the NS-3 simulator. Finally, Section IVconcludes the paper.II. P ROPOSED ALGORITHM
In this paper, we propose the transmission parametersselection algorithm that allows a gNB to select MCS andthe number of transmission attempts for a UE transmittingURLLC packets in uplink.Let us consider a UE served by a gNB. The gNB configuresthe parameters for the UE, i.e., the resources that are availablefor its uplink transmissions and the transmission parameters,i.e., the MCS and the number of transmission attempts K .To change the parameters configuration at the UE, the gNB transmits to the UE the configuration messages that includethe new MCS and the number of transmission attempts. In itsturn, the UE updates the configuration upon the reception ofthe new parameters. The time-frequency resources assignedto the UE are divided into multiple groups called ResourceBlock Groups (RBGs). When the UE transmits a packet, itselects an appropriate number of RBGs corresponding to theselected MCS and the packet size. To avoid correlated errorsin the consecutive RBGs, we assume that the UE selects RBGsuniformly spaced in the frequency domain. These RBGs areselected independently for each attempt. The time slots forrepeated attempts are also selected uniformly spaced withinthe packet delay budget to avoid correlated errors in time.Initially, after the connection establishment, the gNB con-figures the most robust uplink transmission parameters (MCS0 and K max , where K max is the maximum number of trans-mission attempts) because there is no actual SNR statisticsfor this UE. Based on Sounding Reference Signals (SRS)periodically transmitted by the UE, the gNB estimates theaverage SNR for each RBG. Then, for each combination of thetransmission parameters MCS and K ( K = 1 , . . . , K max ), thegNB estimates the BLock Error Rate (BLER), i.e., the errorprobability of a single transmission attempt, as follows: • The gNB calculates the number of RBGs M MCS that isrequired for transmission using the considered MCS. Herewe assume that the gNB has information about the packetsize (e.g, this information can be provided via cross-layerinteraction between the gNB and the URLLC applica-tion [16]). Then, the gNB randomly selects M MCS
RBGsthat are uniformly spaced in the available bandwidth. • We assume that the total power remains the same re-gardless of the number of used RBGs since the UEuses the whole power to transmit data. Hence, the gNBrecalculates SNR according toSNR ( i ) MCS = MM MCS · SNR ( i ) measured , where SNR ( i ) measured is a measured SNR in RBG i , M is atotal number of RBGs in which the UE transmits SRS. • The gNB uses the Exponential Effective Signal-to-noiseratio Mapping (EESM) [17] error model for BLER esti-mation.Let us describe the third step in more detail. Accordingto the EESM model, the gNB maps vector of SNRs for theselected RBGs to a single effective SNR as follows:SNR eff = − β ln (cid:32) | v | (cid:88) n ∈ v exp( − SNR n β ) (cid:33) , where SNR n is the SNR value in the n -th RBG, v is the setof allocated RBGs, and β is a scaling parameter. We use β values obtained in [17].We assume that the gNB uses CC to decode several HARQtransmissions. The effective SNR after q transmission attemptsis calculated as follows:SNR q eff = − β ln | ω q | | ω q | (cid:88) m =1 exp( − β q (cid:88) j =1 SNR m,j ) , ig. 1. SNR-BLER curves for 25 iterations of a decoder. where ω j = { RBG j , ..., RBG j | ω j | } is the set of used RBGsfor j -th transmission attempt (note, that | ω | = ... = | ω q | ),SNR m,j is the SNR experienced in RBG jm .The obtained effective SNR value is mapped to the BLERvalue using the SNR-BLER curves for the correspondingMCS. In this study, we use SNR-BLER curves (Fig. 1)obtained for 25 iterations of a decoder for packet size 32 byteand physical layer parameters described in Section III-A.The packet loss ratio after all K transmission attemptsequals PLR MCS ,K = BLER · . . . · BLER K , where BLER i is the obtained BLER for the i -th transmission attempt.As a result, after each SRS reception, the gNB obtainsPLR estimations for all the configurations { MCS, K } . Foreach configuration, the PLR estimation is averaged with anexponentially weighted moving average as follows: (cid:100) PLR
MCS ,K ( t ) = 1 W · PLR
MCS ,K ( t )+(1 − W ) · (cid:100) PLR
MCS ,K ( t − , where W is the window size, (cid:100) PLR
MCS ,K ( t ) is the averagedPLR estimation after the t -th SRS reception.To avoid frequent reconfiguration, i.e., transmission param-eters changes at the UE, the gNB uses two thresholds, PLR low and PLR high , where PLR low < PLR high . Specifically, the gNBselects the transmission parameters as follows:1) marks all configurations { MCS, K } as valid,if (cid:100) PLR
MCS ,K ( t ) < PLR low , and as invalid, if (cid:100)
PLR
MCS ,K ( t ) > PLR high (see Fig. 2);2) calculates the channel resource consumption M MCS · K for each valid configuration and selects the one thatprovides the minimum channel resource consumption.If the selected configuration differs from the one used bythe UE, the gNB sends the new configuration to the UE.III. P ERFORMANCE EVALUATION
A. Scenario
We evaluate the performance of the proposed algorithm withthe NS-3 simulator [18]. We consider a single gNB and a
Fig. 2. Valid configurations marking.TABLE IS
IMULATION PARAMETERS
Parameter Value
Bandwidth 100 MHz, 16 RBGsSlot length . µs Packet size 32 bytesPacket period 10 msSRS period 5 msUE power 23 dBmUE height 1.5 mgNB height 30 mPropagation model Okumura-Hata model [20]Fading model Extended Pedestrian A (EPA) [21]Simulation time 100000 s K max single UE that generates Constant Bit Rate (CBR) traffic inthe uplink. Specifically, bytes packets are transmitted withperiodicity 10 ms. Each packet should be delivered within the ms time interval with a probability higher than . .SRSs are transmitted every ms.Let us describe physical layer parameters. Following [19],we consider mini-slots (subslots) that consist of two Or-thogonal Frequency-Division Multiplexing (OFDM) symbolscorresponding to control and data channels. The duration ofthe OFDM symbol equals . µs that corresponds to the kHz interval between subcarriers. So, there are 14 slots withinthe packet delay budget. The bandwidth equals MHz andconsists of RBGs [5]. The UE transmission power equals dBm, and it is equally distributed between the selectedRBGs. We summarize all simulation parameters in Table I. B. Analysis of the results
As a characteristic of the average channel quality, whichdecreases with the distance d between the UE and the gNB, weuse the value SNR wb called the wideband SNR. This value iscalculated as follows: SNR wb = ( P T X
P L ( d )) / ( N oise BW ) ,where P T X is a UE transmission power, BW is the availablebandwidth, P L ( d ) is the pathloss, N oise is a noise powerspectral density at the gNB. To model the channel quality ig. 3. Comparison of fixed and adaptive MCS selection. changes in time and frequency domains, we use the ExtendedPedestrian A (EPA) fading model [21].First, let us estimate the reduction of resource consumptionthat can be achieved with the adaptive transmission param-eters selection. For each SNR wb value and each parameterconfiguration { MCS, K } , we carry out the full experimentand consider the PLR and the average number of used RBGsobtained in each experiment. Then, for each SNR wb , weselect the optimal configuration that allows satisfying theURLLC requirements while consuming the minimal numberof RBGs. We compare the resource consumption providedby the proposed adaptive algorithm with the resource con-sumption provided by the optimal configuration and by thefixed selection of the most robust MCS, i.e., MCS 0 (seeFig. 3). The number of transmission attempts K ≤ K max for MCS 0 equals the minimal number that allows satisfyingthe URLLC requirements. We can see that the proposedalgorithm provides more than three times a reduction forthe resource consumption in comparison with the MCS 0selection. Moreover, the results for the proposed algorithmare close to optimal for SNR wb ≥ − dB. Since even withthe most robust configuration the URLLC requirements cannotbe satisfied at SNR wb < − dB, we assume in the furtherexperiments that the UE is located uniformly in the circle withcenter at gNB and radius corresponding to SNR wb = − dB.Now let us study how the parameters W , PLR low andPLR high influence the efficiency of the proposed algorithm. Forthat, we vary the window size W from 500 ms to 20000 msand the thresholds PLR low and PLR high from − to − ,respectively. For each window size value, we find combinationof PLR low and PLR high that provides PLR less than − forall SNR wb values and the minimum RBG usage averaged overSNR wb distribution. Fig. 4 shows how the average numberof used RBGs depends on the window size for the proposedalgorithm. To obtain the optimal and MCS 0 curves, weaverage the number of used RBGs, presented in Fig. 3, overthe SNR wb distribution. According to the results, the number Fig. 4. Channel resource consumption of the proposed algorithm.Fig. 5. Optimal PLR low and PLR high thresholds. of used RBGs for the proposed algorithm converges to optimalwhen the window size increases. Moreover, the proposedalgorithm allows reducing the resource consumption more thantwice in comparison with the fixed selection of the MCS 0.Fig. 5 shows PLR low and PLR high thresholds selected by theproposed algorithm. We can see that PLR high does not dependon the window size and equals − , which correspondsto . URLLC reliability requirement. The thresholdPLR low increases with the window size and tends to PLR high for large window size because the large window allows moreconservatively estimating the packet loss ratio provided bydifferent configurations, and thus it does not require to selectthe lower values for PLR low .According to Fig. 6, the usage of two thresholds allowsthe gNB to rarely reconfigure the transmission parameters. Inparticular, for W = 1 s the parameters should be reconfiguredonce per seconds on average. The reconfiguration frequencyfurther decreases when the window size increases. ig. 6. Stability of the selected transmission parameters. Based on the results presented in Fig. 4 and Fig. 5,we propose the algorithm to select the parameters of thealgorithm. Specifically, we propose to select the windowsize according to the environmental conditions, e.g., the UEmobility parameters, and then select the PLR low according toFig. 5. The threshold PLR high should be set in accordance withthe URLLC reliability requirement.IV. C
ONCLUSION
In this work, we study the problem of uplink transmissionparameter selection with the grant-free channel access. Wepropose the adaptive algorithm for the transmission parametersselection, i.e., MCS and the number of transmission attempts.This algorithm allows satisfying the URLLC requirementswhile reducing the channel resource consumption. The algo-rithm uses the signal-to-noise ratio statistics to estimate thepacket loss ratio for all possible transmission parameter values.Then the algorithm selects the parameter values that requiresthe minimum amount of channel resource while satisfyingthe URLLC requirements. Numerical results obtained with theNS-3 simulator show that the algorithm reduces the channelresource consumption more than twice in comparison withthe fixed robust MCS and optimal K parameters selection.Moreover, with this algorithm, the gNB can select the trans-mission parameters in a long-term perspective, e.g., it canchange the parameters for the time intervals much longer thanthe packet inter-arrival time and, thus, allows reducing theamount of control traffic needed to configure the uplink grant-free transmissions.In this paper, we assume that the gNB allocates dedicatedchannel resources to each UE, and transmission errors are onlycaused by fading. In our future works, we are going to considera scenario in which several UEs can use shared grant-freeresources to transmit their packets. We will adapt the proposedalgorithm to take into account possible interference caused bythe transmission of other UEs in shared resources. R
EFERENCES[1] “5G; Study on scenarios and requirements for next generation accesstechnologies,” 3rd Generation Partnership Project (3GPP), TechnicalReport (TR) 38.913, 2020.[2] H. Ullah, N. G. Nair, A. Moore, C. Nugent, P. Muschamp, andM. Cuevas, “5G communication: an overview of vehicle-to-everything,drones, and healthcare use-cases,”
IEEE Access , vol. 7, pp. 37 251–37 268, 2019.[3] “Framework and overall objectives of the future development of IMT for2020 and beyond,” ITU-R, Recommendation M.2083, September 2015.[4] E. Avdotin, D. Bankov, E. Khorov, and A. Lyakhov, “Enabling massivereal-time applications in IEEE 802.11 be networks,” in
Proceedings ofthe IEEE 30th Annual International Symposium on Personal, Indoor andMobile Radio Communications (PIMRC) . IEEE, 2019, pp. 1–6.[5] “5G; NR; Physical layer procedures for data,” 3rd Generation Partner-ship Project (3GPP), Technical Specification (TS) 38.214, 2020.[6] D. Chase, “Code combining – a maximum-likelihood decoding approachfor combining an arbitrary number of noisy packets,”
IEEE transactionson communications , vol. 33, no. 5, pp. 385–393, 1985.[7] G. Berardinelli, N. Mahmood, R. Abreu, T. Jacobsen, K. Pedersen,I. Kovacs, and P. Mogensen, “Reliability analysis of uplink grant-freetransmission over shared resources,”
IEEE Access , vol. 6, pp. 23 602–23 611, 2018.[8] Y. Liu, Y. Deng, M. Elkashlan, A. Nallanathan, and G. Karagiannidis,“Analyzing grant-free access for URLLC service,”
IEEE Journal onSelected Areas in Communications , 2020, early access.[9] N. Mahmood, R. Abreu, R. Bohnke, M. Schubert, G. Berardinelli, andT. Jacobsen, “Uplink grant-free access solutions for URLLC services in5G new radio,” in
Proceedings of the 16th International Symposium onWireless Communication Systems (ISWCS) . IEEE, 2019, pp. 607–612.[10] T. Jacobsen, R. Abreu, G. Berardinelli, K. Pedersen, P. Mogensen,I. Z. Kovacs, and T. Madsen, “System level analysis of uplink grant-free transmission for URLLC,” in
Proceedings of the IEEE GlobecomWorkshops (GC Wkshps) . IEEE, 2017, pp. 1–6.[11] M. C. Lucas-Estan, J. Gozalvez, and M. Sepulcre, “On the capacityof 5G NR grant-free scheduling with shared radio resources to supportultra-reliable and low-latency communications,”
Sensors , vol. 19, no. 16,p. 3575, 2019.[12] B. Singh, O. Tirkkonen, Z. Li, and M. A. Uusitalo, “Contention-based access for ultra-reliable low latency uplink transmissions,”
IEEEWireless Communications Letters , vol. 7, no. 2, pp. 182–185, 2018.[13] T. Jacobsen, R. Abreu, G. Berardinelli, K. Pedersen, I. Z. Kovacs, andP. E. Mogensen, “System level analysis of K-repetition for uplink grant-free URLLC in 5G NR,” in
Proceedings of the 25th European WirelessConference . VDE, 2019, pp. 96–100.[14] Z. Zhou, R. Ratasuk, N. Mangalvedhe, and A. Ghosh, “Resourceallocation for uplink grant-free ultra-reliable and low latency com-munications,” in
Proceedings of the IEEE 87th Vehicular TechnologyConference (VTC Spring) . IEEE, 2018, pp. 1–5.[15] S. Ozaku, Y. Shimbo, H. Suganuma, and F. Maehara, “Adaptive repe-tition control using terminal mobility for uplink grant-free URLLC,” in
Proceedings of the IEEE 91st Vehicular Technology Conference (VTCSpring) . IEEE, 2020, pp. 1–5.[16] I. F. Akyildiz, E. Khorov, A. Kiryanov, D. Kovkov, A. Krasilov,M. Liubogoshchev, D. Shmelkin, and S. Tang, “XStream: a new platformenabling communication between applications and the 5G network,” in
Proceedings of the IEEE Globecom Workshops (GC Wkshps) . IEEE,2018, pp. 1–6.[17] S. Lagen, K. Wanuga, H. Elkotby, S. Goyal, N. Patriciello, and L. Giup-poni, “New radio physical layer abstraction for system-level simulationsof 5G networks,” in
Proceedings of the IEEE International Conferenceon Communications (ICC)
IEEE Communications Surveys & Tutorials , vol. 20, no. 4, pp. 3098–3130, 2018.[20] T. S. Rappaport,