A QoS harmonization strategy for Wi-Fi and cellular networks convergence

A QoS harmonization strategy for Wi-Fi and Cellular Networks Convergence □Akshay Jain, †Daniel Garcia, †Seyed Mahdi Darroudi, ∗Elena Lopez-Aguilera □Nokia Bell Labs, Finland, †Neutroon Technologies S.L., Spain, ∗Universitat Polit` ecnica de Catalunya, Spain Abstract—Beyond 5G networks will not only present an evolution of the current 5G standards, but they will also provision a path for increased convergence with other Radio Access Technologies (RATs). The Wi-Fi Alliance and 3GPP through their standards releases have presented a lucrative inter-working opportunity for the same. Such a convergence will help to enhance the Quality of Experience (QoE) for the users. However, one of the most important challenge towards a more converged scenario is the provision of uniform Quality of Service (QoS) across cellular and Wi-Fi RATs. Hence in this paper, to the best of our knowledge, we propose the first attempt in literature to homogenize Wi-Fi and 5G QoS. Specifically, a method is presented to map Wi-Fi QoS parameters to 5G QoS Identifiers (5QI). The presented results, which show a clear possibility of the 5G and Wi-Fi QoS harmonization, have been obtained via a real testbed setup. Index Terms—5G, Beyond 5G, 6G, Wi-Fi, Network Slicing I. INTRODUCTION Globally, 5G networks are being rolled out and it is estimated that the number of 5G subscriptions will surpass 1 billion in 2022 [1]. Concurrently, IEEE 802.11 working group most recently released the IEEE 802.11ax standard specification (Wi-Fi 6, 6E) [2], and is currently working to develop the forthcoming IEEE 802.11be (Wi-Fi 7) standard. Moreover, 3GPP and other prominent industrial organizations have also started to develop the 5G and beyond technologies [3]. Given these ongoing activities, the most common question within the research community is : ”What is the next step for wireless technologies?”. While, academics and industries have proposed multiple possible contenders, i.e., non-terrestrial networks (NTNs), Terahertz communications (THz), Visible Light Communications (VLC), etc. [4], one of the most challenging aspect that remains is the convergence of cellular (such as 3G, 4G, 5G) and Wi-Fi networks. Specifically, cellular technologies provision wide area networking capabilities, mobility support, ability to reach difficult (non Line-of-sight) spaces, as well as support for multiple applications and large amounts of users with a high degree of reliability [5]–[7]. However, cellular technologies such as 5G suffer from high capital expenditures (CAPEX), high operational expenditures (OPEX), costly subscription fees and lack of indoor penetrations in the mmWave range. Wi-Fi, on the other hand, is a specialized wireless technology that provides local area access in indoor/outdoor hot-spots and Small Office-Home Office (SOHO) environments. Furthermore, WiFi networks are economical and easy to setup. Given these contrasting differences between both the technologies, it is clear that a converged solution will benefit both technologies and hence improve the user QoE. However, such heterogeneity is also what makes the converged solution a challenging goal. Additionally it is important to state here that, starting from the Wi-Fi 6 and 6E specifications, Orthogonal Frequency Division Multiple Access (OFDMA) mechanism has been adopted. This provisions the capability of granting deterministic and application aware resource allocation [8], similar to the mechanism defined for 3GPP RATs. Furthermore, both 3GPP and the IEEE 802.11 working group have elaborated on the inter-working issue [9], [10]. Consequently, the possibility of cellular networks and Wi-Fi Access Points (AP) working in coordination via mechanisms such as Non-3GPP Inter-working Function (N3IWF), Licensed Assisted Access [11], LTE WiFi aggregation (LWA) [12], etc., has already been explored. However, none of the aforesaid mechanisms and studies have elaborated on provisioning a consistent QoS performance in the event a UE roams between Wi-Fi and cellular RATs. This is due to the existing issues within the Wi-Fi networks, i.e., the best effort mechanisms. Such a harmonization, if provisioned, can improve the QoE for applications that roam between 3GPP and non-3GPP RATs, which in return will also have a positive impact on both public and private network deployments. Hence in this paper, we make a first attempt in literature to present a method that studies the convergence of WiFi and cellular solutions through the determination of a set of equivalent QoS settings in both Wi-Fi and 5G RATs. Consequently, to perform the QoS harmonization study we consider an industrial setup wherein both Wi-Fi and 5G radios coexist as part of a private network. The private network is capable of orchestrating network slices, wherein both Wi-Fi and 5G RATs are part of the same slice. Next, we consider the 5G 5QI models specified in 3GPP specifications [13] and the outputs from a testbed that consists of a Wi-Fi module to demonstrate that such a harmonization between Wi-Fi and 5G is indeed possible for different QoS requirements, thus showing the effectiveness of the method 1. The rest of the paper is organized as follows: In Section II a background regarding QoS in 5G and Wi-Fi is presented. Sec1The reason for considering only the Wi-Fi module for the testbed is that while for 5G the QoS values are specified with a high degree of confidence, for Wi-Fi, due to their best effort mechanism, such QoS numbers do not exist. Hence, we use a testbed to determine the QoS parameters for a Wi-Fi network, which we will describe in more detail in the following sections. tion III describes the method proposed for the QoS mapping and the testbed setup for the experiments performed. Section IV provides a discussion on results. The paper is concluded in Section V. II. BACKGROUND Cellular and Wi-Fi networks utilize fundamentally different mechanisms to guarantee a certain level of QoS and hence, the QoEs are different. In the following subsections we will explore how 3GPP and IEEE 802.11 define the 5G and Wi-Fi QoS, respectively. A. 5G Quality of Service 3GPP, via the specification TS 23.501 [14] defines a series of QoS levels for 5G called 5G QoS Identifiers (5QIs). Specifically, for each 5QI value 3GPP specifies a set of QoS requirements in terms of throughput, reliability, latency, packet loss, etc. This information is used both by the Core network and by the Radio Access Network (RAN). The User Plane Function (UPF) of the Core network applies the necessary traffic shaping to ensure that the specified requirements are satisfied. Concurrently, the RAN utilizes this information within its scheduler for performing resource allocation tasks. We now briefly define the three different traffic classes depending on the 5QI assignment: •Non Guaranteed Bitrate (Non-GBR): No throughput requirement can be set, and it will act as a best-effort flow. A latency requirement is defined. •Guaranteed Bitrate (GBR): A custom throughput requirement can be set, named as Guaranteed Flow BitRate (GFBR). A latency requirement is defined. •Delay-critical GBR: A custom throughput requirement can be set , i.e., GFBR, and a very strict latency requirement is defined. Next, we briefly specify the parameters that are defined via a 5QI assignment: •Resource type: Non-GBR, GBR or Delay-critical GBR •Default priority level: Used by the UPF Core function to manage resource allocation. •Packet Delay Budget (PDB): Maximum one-way delay between UE and the N6 interface of the UPF Core function. At least 98% of the packets of a given QoS flow should satisfy this requirement. •Packet Error Rate: Upper bound for the rate of PDUs that have been processed by the sender of a radio link layer but that are not successfully delivered by the corresponding receiver to the upper layer. •Maximum Data Burst Volume: Denotes the largest amount of data that the 5G RAN is required to serve within a period of PDB. •Default Averaging Window: Represents the duration over which the GFBR and MFBR (Maximum Flow Bitrate) shall be calculated. B. Wi-Fi Quality of Service The QoS provisioning methods used in Wi-Fi are very much related to the MAC mechanism, known as Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA). The medium access within Wi-Fi is decentralized and contentionbased. The devices on the network sense the channel to determine if there is any ongoing transmission. Once the channel is free, each device waits a fixed time known as DCF Inter Frame Space (DIFS). If the waiting period ends and no transmissions were detected, then the device transmits. If the channel is sensed busy, or in the event of a packet collision or frame re-transmission due to errors, a random backoff timer is added to the DIFS waiting time. The Random backoff counter is uniformly chosen inside the interval [0, CW ], where CW is called Contention Window. The backoff timer is decreased as long as the channel is sensed idle, paused when a transmission is in progress, and resumed when the channel is sensed idle again for more than the DIFS. When the backoff timer expires, the station attempts transmission. The CW value starts from CWmin; then it is doubled upon each re-transmission until it reaches CWmax, and the waiting time is recalculated. In this way, the medium access protocol tries to provision a collision free approach for transmission for all devices. With the IEEE 802.11e specification [15], this mechanism is extended to add a new QoS method known as Wi-Fi Multimedia (WMM). The new MAC mechanism, named Enhanced Distributed Channel Access (EDCA), presents four transmission queues (VO, VI, BE and BK) for each WMM device. These queues are known as Access Categories (ACs). Traffic is associated to an AC queue depending on its Differentiated Services Code Point (DSCP) marking, located in the IP header of the packet. Next, each one of these queues acts as an independent Distributed Channel Access (DCA) device. These transmission queues calculate their waiting time independently. This waiting time is composed of a fixed component called Arbitration inter-frame spacing (AIFS) and a randomly calculated component (backoff) in the following way: waiting =AIF S +rand(0, CW )·σ(1) AIF S =SIF S +AIF SN ·σ(2) where σrepresents the slot time and SIFS refers to the Short Interframe Space. The values of AIFSN and CW are variable and dependent on the AC. Note that, the backoff time is only included for frame re-transmissions. Hence in this way, ACs with a more critical traffic will have a higher possibility of obtaining a low waiting time and, therefore, its traffic is prioritized. Once an AC queue obtains the opportunity to transmit it transmits for a period of time defined by the Transmission time Opportunity (TXOP) parameter. In this manner the access is not limited to the transmission of one MAC Service Data Unit (MSDU). The default EDCA parameters are given in Table I. Note that, WMM parameters are distributed by the Wi-Fi AP in the beacon frames [15]. TABLE I DEFAULT EDCA PARAMETERS Access Category AIFSN CWmin CWmax TXOP Voice (VO) 2 7 15 3.264ms Video (VI) 2 15 31 6.016ms Best Effort (BE) 3 31 1023 - Background (BK) 7 31 1023 - III. QOSMAPPING In this section we introduce the 5G and Wi-Fi QoS mapping methodology. 5G QoS specification defines concrete latency values depending on the 5QI [16]. On the other hand, while IEEE 802.11e defines a QoS-provisioning mechanism, it is contention-based and therefore concrete QoS values have not been provided. Henceforth, a method is presented in Section III-A and Section III-B to empirically match 5QI parameters to WMM parameters while maintaining a high degree of QoS reliability. This method was tested in a real testbed, which we will describe in detail in Section III-C. A. Methodology The mapping method proposed in this paper is based on latency measurements. Latency is used instead of throughput as 5G 5QI specification provides concrete latency values, but not concrete throughput values. Some considerations have to be taken into account: •5G specification defines latency requirements (PDB) as one-way UE-to-UPF delay [14]. Note that, the gNB-toUPF delay is also defined within the specification for each 5QI. Hence, the UE-to-gNB one-way delay can be obtained by subtracting this value from the PDB. This can be seen in Figure 1. Fig. 1. Packet Delay Budget computation (X ms is a representative value with different value from UE to gNB and gNB to UPF) on the user plane of a 5G network •This method assumes a symmetrical connection, meaning that UE-to-gNB delay is assumed to be equivalent to gNB-to-UE delay. In the same way, the IEEE 802.11 terminal (STA)-to-AP delay is assumed to be equivalent to AP-to-STA delay. Next, we define step-wise methodology for the proposed QoS mapping as follows: 1) Firstly, the UE-to-gNB one-way delay is obtained for each 5QI following Figure 1 method. 2) Secondly, the UE-to-gNB round trip time (RTT) is obtained by multiplying the one-way delay obtained in step 1 by 2(symmetrical network). 3) In step 3, the STA-AP RTT measurements are obtained from the Wi-Fi testbed and a histogram for the said measurement is computed. 4) Next, a semi-guaranteed STA-AP RTT is calculated by a cut-off at a specific percentile of latency measurements. The procedure for computing the semi-guaranteed latency is presented in Section III-B. 5) Lastly, from all the UE-to-gNB RTT values that are equal or bigger than the value obtained in step 4, the 5QI setting with a latency performance closer to the Wi-Fi scenario is chosen as the equivalent QoS in 5G network. Furthermore, 5QIs with a UE-to-gNB RTT lower than the semi-guaranteed STA-AP RTT are discarded as the required QoS requirements can not be satisfied by the Wi-Fi scenario. Fig. 2. QoS mapping method example Note that, the process is repeated for each AC and set of WMM parameters from step 3 onward. If the obtained latencyequivalent 5QI is GBR or Delay-critical GBR, then the equivalent GFBR can be calculated by obtaining the semi-guaranteed throughput. The computation of the semi-guaranteed throughput is described in Section III-B. We present a flow diagram of the aforementioned method to compute the equivalent 5G and Wi-Fi QoS values in Fig. 2. B. Semi-Guaranteed Throughput and Latency computation Given that the MAC mechanism in Wi-Fi is CSMA/CA based, it is not possible to guarantee a certain QoS . Nevertheless, it is possible to determine values of QoS with a given level of probability. In this sense, a Wi-Fi scenario that satisfies a given throughput requirement with a probability greater than 0.99 can be compared with a 5G scenario serving a GBR flow. We will call this value a semi-guaranteed throughput. This value can be found by applying a threshold to the histogram of the samples obtained through the tests. In the same way, semi-guaranteed latency can be found for each AC and set of WMM parameters. From Figs. 3 and 4, we present scenarios for semi-guaranteed throughput and semi-guaranteed latency determination, respectively. Specifically, in both these scenarios, the requirement values are determined as follows: •Semi-guaranteed throughput: In real 5G scenarios throughput requirements are not satisfied for 100% of the cases due to phenomena like channel shadowing, Doppler effect, etc. Hence, satisfying a throughput requirement for 99% of the cases can be considered a reasonable estimate from a CSMA/CA based technology. Thus, semi-guaranteed throughput for a given Wi-Fi scenario will be one which satisfies the condition of 99% of the measured throughput samples being bigger than a given throughput threshold. In Fig. 3, an example of semi-guaranteed throughput computation is shown. It is computed by finding the throughput value below which 1% of the total number of samples are found. Fig. 3. Semi-guaranteed throughput computation (Average Throughput = 24.35 Mbps; 1 percentile throughput = 20.75 Mbps) •Semi-guaranteed latency: For the purpose of ensuring PDB reliability, a threshold value of 98% is defined within the 5QI specification. Hence, semi-guaranteed latency for a given Wi-Fi scenario will be determined by the latency value above which 2% of the measured latency samples exist. In Fig. 4, an example of semi-guaranteed latency computation is shown. As specified above, it is determined by finding the latency value in histogram above which 2% of the total number of observed samples are found. C. Testbed description As stated earlier, for the computation of the semi-guaranteed throughput and latency values, i.e., Wi-Fi QoS values, a testbed was utilized. Through the testbed, the QoS measures for different WMM scenarios, wherein each WMM scenario is defined by a set of WMM parameters, was computed. For instance, the parameter values in Table I describe a specific WMM scenario. Specifically, the testbed used consists of an AP with a fixed position and an STA placed two meters away. The STA has a RTL8821CE wireless card. The AP is a Gateworks Ventana GW5400 running hostapd. It uses a WLE200NX wireless card. Next, for the throughput tests the Iperf3 tool was used, wherein the AP was acting as the server (uplink) and using the User Datagram Protocol (UDP) to avoid transport layer ACKs. Additionally, to perform the latency tests the ping tool Fig. 4. Semi-guaranteed latency computation (Average latency = 5.31 ms; 98 percentile latency = 14.42 ms) was used, which is readily available in any Linux distribution. Given that multitude of factors can affect the results of the WiFi QoS, our aim with the testbed was to keep the scenario as static as possible. This would consequently allow us to clearly report the impact of WMM parameters on the expected QoS in Wi-Fi networks. Hence, we disabled the mechanisms for Modulation Coding Scheme (MCS) adaptation and Multipleinput Multiple-output (MIMO) via the use of Single-input Single-output (SISO) system. The testbed was deployed in a closed room wherein only the AP and the STA were placed. The location of the transmitters and receivers were not altered throughout the duration of experiments. As a result, this allowed us to achieve throughput and latency values with very low variance. The scenario characteristics are presented in Table II. TABLE II SCENARIO CHARACTERISTICS MCS uplink 7 MCS downlink 7 MIMO type SISO Short GI Disabled Bandwidth 40MHz Channel number 149 Power received by station -55dBm It is important to state here that, the MCS is fixed to 7 as the power received by the AP was good enough to demodulate the symbols with a very low packet loss rate. Moreover, UDP is used to avoid transport layer ACKs produced by TCP. Additionally, the channel number was determined after scanning the environment for a channel with low interference values. Next, the throughput and latency samples were obtained for each Access Category and for nine different WMM scenarios. However, due to space constraints we only present the results for three WMM scenarios in the next section. Further, to force contention between ACs, tests were performed with all four AC queues attempting for simultaneous transmissions. Lastly, TABLE III WI-FIWMM PARAMETERS BASED QOSTO 5QI MAPPING AC WMM Parameters RTT latency (98%) Throughput (99%) GBR 5QIs non-GBR 5QIs VO CW 3-7, AIFSN 2, TXOP 1504 µsec 11.0090 ms 25.18 Mbps 82,83,88 80 VI CW 7-15, AIFSN 2, TXOP 3008 µsec 11.1646 ms 23.90 Mbps 82,83,88 80 BE CW 15-1023, AIFSN 3 12.8751 ms 8.90 Mbps 82,83,88 80 BK CW 15-1023 AIFSN 7 14.2055 ms 5.55 Mbps 82,83,88 80 VO CW 3-7, AIFSN 2, TXOP 1504 µsec 11.8619 ms 23.06 Mbps 82,83,88 80 VI CW 7-15, AIFSN 2, TXOP 3008 µsec 11.6956 ms 22.51 Mbps 82,83,88 80 BE CW 15-1023, AIFSN 3 18.2485 ms 8.46 Mbps 89 79 BK CW 3-7, AIFSN 7 13.6458 ms 6.72 Mbps 82,83,88 80 VO CW 3-7, AIFSN 2, TXOP 1504 µsec 16.0735 ms 25.50 Mbps 82,83,88 79 VI CW 7-15, AIFSN 2, TXOP 3008 µsec 19.2077 ms 24.32 Mbps 89 79 BE CW 15-1023, AIFSN 3 29.6581 ms 9.56 Mbps 90 79 BK CW 15-1023 AIFSN 7, TXOP 12032 µsec 24.2720 ms 6.36 Mbps 89 79 the sampling rate for measurements in the throughput tests for each AC was 10 Hz during 300s, while for the latency tests it was 5 Hz (100 ping tests in 20 seconds) for each AC. IV. RESULTS AND DISCUSSION We now present a discussion of the obtained results in this section. Table III presents a mapping of the Wi-Fi QoS (i.e., ACs), given different WMM parameters and the obtained 98% RTT latency (semi-guaranteed latency) and 99% throughput (semi-guaranteed throughput) for each AC, with the 5G 5QI values. Note that the WMM parameter values mentioned in the 2nd column of Table III consist of the different values of CW, AIFS and TXOP used for the experiment. Additionally, the first four results in Table III utilize the default WMM parameter values, whilst results from rows 5-8 are computed by using different CW values. Next, the last four results are generated by varying the TXOP WMM parameter. From the results, we can observe that for each of the ACs we have been able to determine the equivalent GBR and non-GBR 5QIs. For instance, in the case of the VO AC, where we vary the CW parameter (row 5 in Table III), the results for throughput and latency are 23.06 Mbps and 11.8619 ms, respectively. When we observe the throughput and PDB requirements for 5QI values in [16], it can be observed that GBR 5QI 82, 83 and 88 and non-GBR 5QI 80 have corresponding QoS requirements. Hence, this shows the validity of our proposed method. Similarly, using our proposed approach, mappings for the other ACs to their corresponding 5QIs have been listed in Table III. V. CONCLUSION Based on latency and throughput measurements, a first approach in literature has been presented to find a mapping of the observed QoS in Wi-Fi to an equivalent 5G QoS, i.e., 5QI. The results show the possibility of a Wi-Fi and 5G QoS harmonization despite its fundamental differences. It must be stated that the present work assumes certain ideal conditions, and a real world scenario may impact the mapping presented. Hence, as part of future work we plan to utilize machine learning techniques to predict equivalent 5QI for a given WiFi scenario. Additionally upcoming Wi-Fi standards will also be explored as part of the future work. ACKNOWLEDGEMENTS This research was funded in part by the Spanish MCIN/AEI/10.13039/501100011033 through project PID2019-106808RA-I00. REFERENCES [1] Ericsson, “Ericsson mobility report - june 2022.” [Online]. Available: https://www.ericsson.com/49d3a0/assets/local/reports-papers/ mobility-report/documents/2022/ericsson-mobility-report-june-2022. pdf [2] IEEE, “Ieee standard for information technology–telecommunications and information exchange between systems local and metropolitan area networks–specific requirements part 11: Wireless lan medium access control (mac) and physical layer (phy) specifications amendment 1: Enhancements for high-efficiency wlan,” pp. 1–767, 2021. [3] M. Hoffman et al., “D1.1: 6g vision, use cases and key societal values,” 2021. [Online]. Available: https://hexa-x.eu/wp-content/uploads/2021/ 02/Hexa-X D1.1.pdf [4] A. Jain, E. Lopez-Aguilera, and I. Demirkol, “Are mobility management solutions ready for 5g and beyond?” Computer Communications, vol. 161, pp. 50–75, 2020. [5] ——, “Evolutionary 4g/5g network architecture assisted efficient handover signaling,” IEEE Access, vol. 7, pp. 256–283, 2019. [6] ——, “User association and resource allocation in 5g (aura-5g): A joint optimization framework,” Computer Networks, vol. 192, p. 108063, 2021. [7] A. Dogra, R. K. Jha, and S. Jain, “A survey on beyond 5g network with the advent of 6g: Architecture and emerging technologies,” IEEE Access, vol. 9, pp. 67 512–67 547, 2021. [8] S. Chauhan et al., “Ieee 802.11be: A review on wi-fi 7 use cases,” in 2021 9th Intl. Conf. on Rel., Inf. Tech. and Optim. (ICRITO), 2021, pp. 1–7. [9] 3GPP, “Ts 23.501 version 17.6.0 release 16, section 4.2.8 ”support of non-3gpp access”.” [10] IEEE, “802.11 working group aani standing committee, on the interworking between 3gpp 5g network and wlan.” [Online]. Available: https://www.ieee802.org/11/Reports/aani update.html [11] N. Rupasinghe and I. Guvenc, “Licensed-assisted access for wifi-lte coexistence in the unlicensed spectrum,” in 2014 IEEE Globecom Workshops (GC Wkshps), 2014, pp. 894–899. [12] R. Bajracharya et al., “Lwa in 5g: State-of-the-art architecture, opportunities, and research challenges,” IEEE Communications Magazine, vol. 56, no. 10, pp. 134–141, 2018. [13] 3GPP, “Ts 23.501 version 17.6.0 release 16, section 5.7.3 ”5g qos characteristics”.” [14] ——, “Ts 23.501 version 17.6.0 release 16, ”system architecture for the 5g system (5gs)”.” [15] W. Alliance, “Wmm specifications.” [Online]. Available: https: //www.wi-fi.org/discover-wi-fi/wi-fi-certified-wmm-programs [16] 3GPP, “Ts 23.501 version 17.6.0 release 16, section 5.7.4 ”standardized 5qi to qos characteristics mapping”.”