Kai Kang , Zhenghang Zhu , Dehua Liu , Wuxiong Zhang , Hua Qian＊

1 Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai 201210, China

2 Shanghai Research Center for Wireless Communications, Shanghai 201210, China

* The corresponding author, Email： qianh@sari.ac.cn.

I. INTRODUCTION

Wireless data traffic grows exponentially in recent years. Among all wireless payloads,more than 70% occurs indoor now [1]. Wireless local area network (WLAN) or Wireless Fidelity (Wi-Fi) plays an important role for indoor wireless communications. New technologies are researched or adopted by Wi-fimore quickly than ever to improve the throughput and reliability. Recent advances include multi-user multiple-input multiple-output(MU-MIMO), orthogonal frequency division multiple access (OFDMA), and etc. Wi-fiis also recognized as an important part in the fifth generation (5G) evolution. Heterogeneous networks with both cellular access and Wi-fiaccess are typical in dense areas[2]. An open Wi-fiplatform is preferred for fast new technologies veri fication.

There are several published works on Wi-fisystem. The software radio (Sora) developed by Microsoft is one of the pioneer platforms in advancing the state of the art in software de fined radio (SDR) technology [3]. This platform enabled numerous researches in software de fined radios. On the other hand, its capability is limited, which can only support IEEE (Institute of Electrical and Electronics Engineers)802.11 a/g/n and up to 40 MHz bandwidth.This platform is not able to catch up with the next generation Wi-firesearch [2]. The wireless open-access research platform (WARP)initiated by Rice University is a scalable and extensible programmable wireless platform,built from the ground up to prototype advanced wireless networks. WARP combines high-performance programmable hardware with an open-source repository of reference designs and support materials. However, the WARP is running on the Field Programmable Gate Array(FPGA) based hardware, which is still not user friendly for general developers[4]. A virtual Wi-fitest-bed is employed in University of Thessaly, Greece, which designs a virtual AP enabled test bed by employing multi-service set identifier (SSID) services.This open test-bed can support large scale experimentally networking research, but may not support the Wi-firesearch on the physical layer [5].

Other open platforms for wireless communications include the Open5GCore and etc.,which can also serve as a quick prototype to support new researches in related areas [6].High throughput and additional flexibility are always the focus.

Wi-fiis an irreplaceable part on the subject of ubiquitous wireless connectivity towards building the new 5G ecosystem. In this paper,an open Wi-fiplatform is introduced. This Wi-fiplatform is based on the general-purpose processor and universal software radio peripheral (USRP), which supports IEEE 802.11a/g/n/ac physical layer and distributed coordination function (DCF) media access control (MAC) mechanism. The system is optimized for real-time signal processing in general purpose processors. It is also capable of supporting the upcoming IEEE 802.11ax standard. Comparing to other Wi-fiplatforms mentioned above, our open Wi-fiplatform can support the latest Wi-fistandard, IEEE 802.11ac, which features up to 160 MHz bandwidth. In addition, both MAC and PHY implementations are developed in software.This open Wi-fiplatform helps to accelerate the research on the future Wi-fiphysical layer and MAC layer and help to carry out cellular and Wi-finetwork convergence experiments.

The rest of this paper is organized as follows. Section II introduces the Wi-fiphysical layer implementations, Section III specifies the Wi-fiMAC layer implementations, Section IV gives the details on open Wi-fiplatform hardware. Numerical and experimental results are given in Section V. Finally, Section VI concludes this paper.

II. IEEE 802.11AC PHY SYSTEM AND IMPLEMENTATION

In Wi-fisystems, there are strict timing requirements for the frame intervals. For example, the distributed inter-frame space (DIFS)time in IEEE 802.11ac is 34μs. The ACK frame has to be sent within the DIFS interval.

On the other hand, for the open Wi-fiplatform, all the processing is implemented in general-purpose processors. Due to limited processor’s frequency and limited number of cores, it is challenging to complete the receiving processing within the prescribed time.

In this section, the Wi-fiphysical layer,especially the IEEE 802.11ac, is introduced.The transmitter and receiver architectures are shown in detail. We also discuss the real-time processing consideration on key components in the system.

2.1 IEEE 802.11ac PHY introduction

IEEE 802.11ac is the fifth generation of Wi-fistandard, which was officially published in December 2013. IEEE 802.11ac has several characters over past generations of Wi-fito support high throughput [7, 8].

– Extended channel bandwidth:mandatory 80 MHz channel bandwidth, and optional 160 MHz, 80+80 MHz channel bandwidth;cf. 40 MHz maximum in 11n.

– Higher order modulation:256 quadrature amplitude modulation (QAM), with rate 3/4 and 5/6 coding, added as optional modes(vs. 64 QAM, rate 5/6 coding maximum in 11n).

– More MIMO spatial streams:support for up to eight spatial streams. (vs. four in 11n)

– Downlink MU-MIMO:allows up to four simultaneous downlink MU-MIMO clients.Multiple stations, each with one or more antennas, receive independent data streams simultaneously.

When 160 MHz bandwidth, 256 QAM modulation, rate 5/6 coding and short guard interval (GI) are employed, the 11ac physical layer can support up to 6.93 Gbps data rate.

IEEE 802.11ac use OFDM, whose subcarrier spacing is 312.5 kHz. For example, in the case of 80 MHz bandwidth, there are 256 subcarriers, including 234 data subcarriers, 8 pilot subcarriers and a few null subcarriers. An OFDM symbol duration is 4.0μs with normal GI, or 3.6μs with short GI.

Fig. 1 shows the 802.11ac physical layer frame format.

The non-high throughput (non-HT) short training field (L-STF) includes 10 identical short training symbols (each is 0.8μs), used for automatic gain control (AGC) and coarse synchronization at the receiver. The non-HT long training field (L-LTF) includes two long training symbols with a 1.6μs special GI at the beginning, used for fine synchronization and channel estimation. The non-HT SIGNAL field(L-SIG) contains the non-HT RATE and the non-HT LENGTH fields. The L-STF, L-LTF and the L-SIG are the same as the fields in IEEE 802.11a/g/n for backward compatibility.These three fields are called the non-very high throughput (non-VHT) portion. In MIMO scenarios, the same signals are transmitted on all antennas.

The VHT signal A (VHT-SIG-A) field consists of two OFDM symbols, which carry the necessary information required to interpret this frame, including bandwidth, number of spatial streams used, group ID, modulation and coding scheme (MCS) (only for single user(SU) frames), coding and etc. The VHT short training field (VHT-STF) field is similar to the L-STF, except that it only contains 5 short training symbols. Its main purpose is to improve AGC estimation in a MIMO transmission. The VHT long training field (VHT-LTF)field is a fixed sequence, providing a means for the receiver to estimate the MIMO channel. The number of OFDM symbols in this field depends on the number of spatial streams used in this frame. The VHT signal B (VHTSIG-B) field includes the data length and the MCS information (only for MU frames). The Data field is the scrambled, coded and modulated MAC frame from upper layer. The VHTSIG-A, VHT-STF, VHT-LTF and VHT-SIG-B are unique fields in 802.11ac, which are called the VHT portion.

Convolutional code is used for the L-SIG,VHT-SIG-A, VHT-SIG-B to combat the noise and channel fading, which is modulated by fixed binary phase shift keying (BPSK). Convolution code and optional low density parity check code (LDPC) are used for the data field.The Data field is modulated using BPSK,quadrature phase shift keying (QPSK), 16 QAM, 64 QAM, or 256 QAM [7].

2.2 IEEE 802.11 PHY architecture

Fig. 2 shows the transmitter implementation for the Data field.

The MAC frame and corresponding transmission parameters are received from the MAC layer. First of all, the PHY pad bits and tail bits are appended to the MAC frame with zero bits (at least 6 bits) so that the resulting length is a multiple ofNDBPS(number of data bits per OFDM symbol). The scrambler, initiated with a nonzero seed, generates a pseudorandom scrambling sequence (a 127 bits cycle sequence). The scrambling sequence is then applied to the incoming data to reduce the probability of long consecutive of 0’s or 1’s. The scrambled bits are divided among encoders by sending bits to different encoders in a round robin manner. The number of encoders is determined by rate-dependent parameters. Then the coded bits are interleaved and mapped to BPSK, QPSK, 16 QAM, 64 QAM, or 256 QAM constellation points. The pilots are inserted at the fixed subcarriers. Cyclic shift diversity (CSD) is applied for each space-time stream to avoid the unintentional beamforming. Vectors of constellation points from all space-time streams are expanded via matrix multiplication to produce the input to transmit antennas for beamforming. An inverse discrete Fourier transform (IDFT) module transforms the constellation points into the time domain signal. A normal GI or a short GI is inserted into the beginning of every OFDM symbols. The signal with GI is windowed for better spectral decay, and is sent to the digital to analog converter (DAC) and the radio frequency (RF) module.

Fig. 1 IEEE 802.11ac physical layer frame format [6]

Fig. 3 shows the receiver architecture of the Wi-fisystem. The synchronization module is applied to search for incoming frames.If a frame is detected, according to the cyclic characteristics of the L-STF, the autocorrelation is used for coarse time and frequency synchronization. Based on the coarse synchronization results, cross-correlation is applied to the L-LTF for the fine time and frequency synchronization. After frequency offset compensation, which is implemented as a mixer in time domain, the discrete Fourier transform(DFT) module transforms the time domain signal into frequency domain signal. A single input single output (SISO) channel estimation module estimates the channel based on L-LTF signal for every subcarrier. The equalization module follows to compensate for the channel effects. L-SIG and VHT-SIG-A are decoded to obtain necessary parameters for this frame.

Based on the parameters decoded, a MIMO channel estimation is applied on the VHT-LTF.The VHT-SIG-B and Data fields are equalized and then decoded. The acquired MAC frame and corresponding parameters are sent to the MAC layer for further processing.

2.3 Acceleration of MIMO detection

MIMO is an important feature in wireless communications to increase the physical layer throughput and/or improve the robustness of the transmission. IEEE 802.11n/ac and the forthcoming IEEE 802.11ax are equipped with MIMO. The MIMO detection module at the receiver, which detects different space streams from different antennas, is a key module in the open Wi-fiplatform in terms of complexity and performance trade-off [9, 10].

Fig. 2 Transmitter block diagram for the Data field of a VHT SU PPDU with convolutional code [6]

Zero forcing (ZF) algorithm for MIMO detection is easy to implement. However, this algorithm suffers from noise enhancement and its performance is limited [11]. Minimum mean squared error (MMSE) detection reduces the noise enhancement. However, MMSE is still not an optimal MIMO detection algorithm and its performance is not satisfactory [11].

Fig. 3 IEEE 802.11ac receiver block diagram

On the other hand, the complexity of the ML detection increases exponentially with the increase of the number of spatial streams and the modulation order. For 256 QAM modulation with 2 spatial steams, ML detection needs to search 216(65536) times, which is technically impossible to be implemented in real time in both hardware and the open Wi-fiplatform.

Sphere decoding needs to choose the searching radius properly thus avoids exhaustive search. It includes the following steps.

1) QR factorization of the channel matrixH,

whereRis anm×mupper triangular matrix,andQ=[Q1Q2] is ann×northogonal matrix.Substituting (3) into (2), we have

Searching in sphere decoding can then ex-pressed as a tree, shown in Fig. 4. For a given radiusd, the eligiblesmcan be determined by omitting the second items and beyond, in the right side of the equation (6); then, for the given radius and the givensm, eligiblesm-1can be determined. Go along all the eligible lattices,s, can be found. The transmit vectorswith the minimum distance is the output of the sphere decoding.

The choice of radiusdgreatly affects the complexity of sphere decoding. In our implementation, a suboptimal MMSE detection is applied first, a radius extracted from the distance between the output of MMSE and the received signalxis applied as the initial radius, which guarantees at least a solution within this radius. During the tree search, the radius is reduced when a lattice with less distance is found. This operation can reduce the number of the search times greatly.

Applying the simplified sphere decoding,the search times is greatly reduced for real-time application.

2.4 Acceleration of viterbi decoding

In Wi-fisystem, to combat for the channel fading and burst error, (2, 1, 7) convolutional code is mandatory. A sequence withnbits is encoded into 2n+12 bits with convolutional code. Viterbi decoding is the maximum likelihood sequence decoding algorithm for convolutional code, which has a pipeline structure and is well suited for hardware implementation. However, latency of the Viterbi decoder is a major concern for real-time processing[14].

The radix-2 Viterbi decoding is a standard approach for the rate 1/2 convolutional code.During each time interval, the decoding module receives a pair of channel symbols and computes the distance metric accordingly. An example for radix-2 state transition diagram is shown in Fig. 5(a). The latency of the radix-2 Viterbi decoder is determined by the length of the input data [14].

To take advantage of multiple cores in the processor and accelerating the decoding process, radix-4 Viterbi decoding can be applied.In this case, four channel symbols are process at one time and the distance metrics are calculated in parallel in different processor cores.For one state, a surviving path can be found from the four accumulated error metrics, as shown in Fig. 5(b). The latency of the Viterbi decoder is cut to a half, which is crucial to meet the DIFS timing requirement [15].

Fig. 4 Search tree for the sphere decoding

III. IEEE 802.11MAC SYSTEM AND IMPLEMENTATION

IEEE 802.11 MAC layer’s function includes framing, channel management, connection management, Quality of Service (QoS), link adaptation, security and etc.

In open Wi-fiplatform, the basic MAC functions are implemented. Key parameters can be adjusted as needed. In addition, a signaling interface is designed for possible cellular and Wi-ficonvergence network.

3.1 IEEE 802.11 MAC frame structure

The MAC frame format contains a set offields that occur in a fixed order in all frames. Fig. 6 depicts the general MAC frame format [7]

The Frame Control field consists of the following subfields： Protocol Version, Type,Subtype, To DS, From DS, More Fragments,Retry, Power Management, More Data, Protected Frame, and Order.

The Duration/ID field is 16 bits in length,indicating this frame’s duration.

There are four address fields in the MAC frame format. These fields are used to indicate the basic service set identi fier (BSSID), source address (SA), destination address (DA), transmitting station address (TA), and receiving station address (RA).

The Sequence Control field is 16 bits in length and consists of two subfields, the Sequence Number and the Fragment Number.

The QoS Control field is a 16-bit field that identi fies the various QoS-related information.

The Frame Body is a variable-length field that contains information speci fic to individual frame types and subtypes.

The frame check sequence (FCS) field is a 32-bit field containing a 32-bit cyclic redundancy check (CRC). The FCS is calculated over all the fields of the MAC header and the Frame Body field.

The FCS is calculated using the following standard generator polynomial of degree 32：

Fig. 5 (a) State transition diagram in Radix-2 Viterbi decodig

Fig. 5 (b) State transition diagram in Radix-4 Viterbi decodig

3.2 IEEE 802.11 DCF

The basic channel access method of the IEEE 802.11 MAC is DCF known as carrier sense multiple access with collision avoidance(CSMA/CA).

For a station to transmit, it shall sense the medium to determine if there is any other station transmitting. If the medium is available,the transmission may proceed. The CSMA/CA distributed algorithm mandates that a gap of a minimum speci fied duration exists between contiguous frame sequences. A transmitting station shall verify that the medium is idle during this period before attempting to transmit.

Fig. 6 IEEE 802.11 MAC frame format [6]

If the medium is busy, the station shall defer until the end of the current transmission. After deferral, or prior to attempting to transmit again immediately after a successful transmission, the station shall select a random backoff interval and count down the backoff interval counter while the medium is idle. A transmission is successful either when an ACK frame is received from the station addressed by the RA field of the transmitted frame or when a frame with a group address in the RA field is transmitted completely. A refinement of the method may be used under various circumstances to further minimize collisions. In this case, the transmitting and receiving station exchange short control frames (RTS and CTS frames) after determining that the medium is idle and after any deferrals or backoffs, prior to data transmission [16].

3.3 Signaling interface with 5G

5G network will be a heterogeneous network,including cellular network and Wi-finetwork.The wireless resource will be allocated according to the User Equipment’s (UE) location and requirement, to satisfy the UE’s QoS.Many studies have been carried out in this area. Since the CSMA/CA based distributed competitive mechanism cannot achieve the resource allocation and scheduling, cellular assisted scheduling is reasonable. In this platform, a signaling interface between the 5G core network and Wi-fiAP is de fined and developed [17].

This signaling interface achieves the following functions： 1, UEs or Wi-fistations apply for wireless resource; 2, APs report their current load status to 5G core network; 3, 5G core network sends connection admission control (CAC) signaling to the AP to allow given UEs getting the access to the AP; 4, 5G core network sends resource allocation scheme to the AP.

To provide wireless resource allocation in the Wi-Fi, two mechanisms are designed.

1, Access control. Only allowed users could access to the Wi-fiAP, which ensure the connected users’ QoS.

2, At the MAC layer of the AP, for the downlink data, one independent queue is used for a user’s data packet instead of one queue supporting all users. When the AP obtains a transmission opportunity by competition, a given user’s data packet is sent according to the resource allocation scheme to ensure the users’ QoS, as shown in Fig. 7.

The proposed resource allocation can allocate the wireless resource for the downlink Transmission Control Protocol (TCP) or User Datagram Protocol (UDP) data stream directly.For the uplink TCP data stream, this resource allocation scheme can also directly control the downlink ACKs’ delay, and then trigger the congestion control in TCP to adjust the uplink transmission rate, to achieve the efficient uplink resource allocation.

Fig. 7 Resource allocation in open Wi-fiMAC

IV. HARDWARE IMPLEMENTATION

In this section, we discuss the open Wi-fihardware architecture in detail.

4.1 System setup

The open Wi-fiplatform is based on a USRP radio front end, and a server, as shown in Fig.8.

The server’s model is HP ML350 Gen9,with two Intel?Xeon?E5-2620 v3 processors,64 GB DDR4 memory and three 600 GB hard disks. The processor’s frequency is 3.2GHz,with total 12 cores supporting 24 threads. The operation system is the low latency Ubuntu Linux system. The server sends baseband digital signal to and receives from the USRP radio front end through two 10 Gbps Ethernet cable at 160 MHz sample rate. The interface between the AP and the 5G core network is carried over Ethernet cable. The server receives the resource manage control instruction from the 5G core network module and sends the AP load status report to 5G core network.

Fig. 8 Open Wi-fiPlatform

Fig. 9 One X310 and two UBX160

In the AP’s server, MAC functions such as authentication, resource manager, MAC frame generation and resolution are implemented.Only the authenticated UE could access to this AP. Downlink IP packets are scheduled according to the 5G core network radio resource management instructions. IP packets are added with MAC header and FCS. When the channel is idle or its backoff counter reaches zero, the MAC layer sends the frame to physical layer with MCS, frame format, number of space streams, transmitting power and other parameters.

In the physical layer processing, when a MAC frame is received from the MAC layer,this frame is scrambled, coded, interleaved,modulated, inserted with pilots. The packet is inserted with PLCP header and converted to time domain signal. The 16-bit time domain in-phase and quadrature (IQ) baseband signals are transmitted to X310 through direct memory access (DMA).

Fig. 9 shows the USRP module. This module is the X310 and UBX160 boards from Ettus Research Company. The Ettus Research USRP X310 is a high-performance, scalable SDR platform. In addition to providing bestin-class hardware performance, the open source software architecture of X310 provides cross-platform UHD driver support making it compatible with a large number of supported development frameworks, reference architectures, and open source projects.

At the heart of the USRP X310, the XC7K410T FPGA provides high-speed connectivity among all major components within the device including radio frontends, host interfaces, and DDR3 memory. The default FPGA core included in UHD provides all of the functional blocks for digital down-conversion and up-conversion, fine-frequency tuning, and other digital signal processor (DSP)functions allowing it to be interchangeable with other USRP devices using the UHD architecture. The 160 MHz sampled signal from the server is filtered and then sent to the two UBX160 daughterboards, respectively [18,19].

The UBX 160 daughterboard is a full-duplex wideband transceiver that covers frequencies from 10 MHz to 6 GHz with up to 160 MHz of instantaneous bandwidth. Coherent and phase-aligned operation across multiple UBX daughterboards enables MIMO applications. The 160 MHz sampled digital signal from X310 is converted into analog signal by DAC, and then up converted to 2.4GHz or 5GHz RF signal [20].

The receiving process is just opposite to the transmitting process as described above.The UBX160 converts the RF signal into 16-bit 160 MHz digital sampled IQ baseband signal, and then transfer the signal to X310.The digital signal is filtered and synchronized. If a physical layer header is detected,the synchronized 16 bits digital IQ signal is sent to the server by DMA. The headers are decoded and analyzed to get the necessary parameters. After data field is demodulated and decoded in the physical layer, a MAC frame is obtained and sent to the MAC layer with the receiving parameters. The MAC layer analyzes the MAC frame and checks the FCS.If this frame’s destination address matches the local address and the FCS check is correct, an ACK frame is sent to confirm the reception and the data is sent to upper layer. Because the CSMA/CA mechanism is time division duplex(TDD), so at a given time, an AP or a station can only be in a state of transmitter, receiver or idle. The hardware resources can be reused during the transmitting and the receiving.

4.2 Capability and resource usage

The open Wi-fiplatform supports IEEE 802.11ac physical layer and is backward compatible with IEEE 802.11a/g/n, in 2.4G and 5GHz band. 20 MHz, 40 MHz, 80 MHz and 160 MHz bandwidth are supported. Both APs and stations have 2 antennas to support 2*2 MIMO transceiver. Specific parameters are shown in table I.

C language is used to develop the IEEE 802.11ac physical layer and MAC layer system. The total code length is about 120,000 lines. All threads in one CPU are used to process the transmission. For IEEE 802.11ac, 80 MHz bandwidth and MCS 3 with single antenna, during the system running, the overall CPU occupancy is not higher than 50%, and the used thread occupancy is close to 100%,the maximum memory usage is about 8GB.

V. EXPERIMENTS

In this section, a set of experiments are carried out using this open Wi-fiplatform.

5.1 Point to point transmission

The first experiment is to test the functionality of the proposed open Wi-fiplatform.A scenario with one AP and one station is established in a 4m×5m laboratory. The AP and station are located on the opposite wall,a distance of 5m. The AP is programmed in 20 MHz IEEE802.11ac mode with 1 antenna.The AP continually transmits packets with 1000 octets payload with varying transmission power,while the station receives the packets and calculates the packet error ratio (PER). By measuring the received signal strength indication (RSSI), we conclude that the path loss is about 65 dB.

Table I Speci fic parameters of open Wi-fiplatform

Fig. 10 The simulated and actual test results

In Fig. 10, the solid lines show the PER performance of the signal transmitted with MCS0 (rate 1/2, BPSK); the dashed lines show the PER performance of the signal transmitted with MCS3 (rate 1/2, 16 QAM). The curves with circles represent simulated results with AWGN channel. The curves with cross symbols represent simulated results with Channel A model de fined in the speci fication [21]. The curves with triangle symbols represent the measurement results in the lab tests. From this figure, we observe that with the increase of modulation order, the demand for transmission power or SNR is increased, which agrees with the design principle of the system to adaptively adjust data rate according to the channel condition. For a given MCS, the lab test performance sits in between the simulation results of the AWGN channel and the Channel A model. Channel A model is a typical office environment with 18 taps. In our lab experiment,although it is also an office environment, the line-of-sight (LOS) path between the AP and the station is strong, thus achieves better performance than that of Channel A model [22].

In the second experiment, we test the network in a multi-user scenario. 1 AP and 4 stations constitute an infrastructure Wi-finetwork in the same laboratory. AP is in the center of the room and the stations are located next to the room wall. Each station requests the online high de finition (HD) video and file download from the server at the same time.

All AP and stations are programmed in 80MHz bandwidth 802.11ac mode with 2×2 MIMO, which ideally supports a baseband rate up to 702 Mbps.

Table II PHY processing time for one OFDM symbol with 2*2 MIMO

5.2 MU-MIMO test

With MU-MIMO, the AP can simultaneously transmit multi-spatial streams to the several stations, respectively. Channel estimation and MU-MIMO beamforming are employed to avoid the interference between stations. Thus,the overall throughput can be improved.

In this experiment, the AP is implemented with a FPGA based Wi-fitest-bed with 4 antennas, which are arranged in a row with the distance of 20cm each. The two stations are based on open Wi-fiplatforms with 2 antennas each.

In the same 4m×5m laboratory, the AP is located in the center of the laboratory. The AP is programmed in 80 MHz bandwidth 802.11ac mode with MCS2 and 4 spatial steams. The stations are programmed with 2 spatial steams.The experiments are summarized in Table III.

The transmission signal of one antenna at the AP is shown in Fig.11. Fig.11 (a) presents the time-domain waveform of a frame. The IQ data are 16 bits signed number with 160 MHz sample rate. The first 1280 data with periodicity is L-SIG, and then the L-LTF follows. Fig.11 (b) shows the power spectral density of the transmitted signal, the transmitted signal’s bandwidth is about 76.5 MHz.

When the two stations are located on the opposite wall, if the MU-MIMO beamforming is applied, the throughput is 15.6 Mbps,which gets 39% throughput improves over the SU-MIMO scenario with beamforming. This throughput improvement is less than 100%due to the channel reporting and MU-MIMO signaling overhead.

When the two stations are located on the same wall, if the MU-MIMO beamforming is applied, the throughput is 10.1 Mbps, a little less than the SU-MIMO scenario with beamforming. This result indicates that the channel orthogonality of the two users’ channel has impact on the performance of the MU-MIMO beamforming. When the two users are located nearby, the orthogonality between the two users’ channel matrices is poor, which results in the inter-user interference.

For any station placement situation, if the MU-MIMO beamforming is not applied, the throughput is near to zero. This indicates that the beamforming at AP does work to fight against inter-user interference.

5.3 Network coexistence test

In this experiment, we would like to use the open Wi-fiplatform to test the future 5G+Wi-finetwork convergence scenario. A server is used to simulate the 5G core network.The signaling of Wi-firesource allocation is transmitted through the interface between the simulated 5G core network and AP, as shown in Fig. 12. The AP and 4 stations are programmed in 80 MHz bandwidth 802.11ac mode with 2 spatial steams [23, 24].

When the simulated 5G core network sends the CAC signaling to the AP, to inform the AP that only station 1 and station 2 can access to the Wi-finetwork. After the stations are powered on, station 1 and 2 access to the network successfully, and the station 3 and 4 can not complete the authentication and association process, and cannot access to the Wi-finetwork.

Table III MU-MIMO experimental results

Fig. 11 a) Tramsmitted time-domain waveform; b) its power spectral density

Fig. 12 The open Wi-Fi’s architecture

When the simulated 5G core network send a resource allocation signaling to the AP, indicating that station 1’s assigned bandwidth is 250 Mbps, station 2’s assigned bandwidth is 100 Mbps and the remaining available bandwidth is evenly allocated between station 3 and station 4. AP implements the allocation of resources by providing separated queues for each station’s packets and coordinating the transmission according to the resource allocation signaling. This test shows that station 1’s throughput fluctuates between 240 and 250 Mbps, and the station 2’s throughput fluctuated between 93 and 100 Mbps. The stations 3 and 4’s throughput varies from 20 to 40 Mbps.

When the AP receives a resource allocation signaling from the simulated 5G core network,indicating that the station 2’s assigned bandwidth is changed from 100 Mbps to 50 Mbps,we observe that station 1’s throughput is still between 240 and 250 Mbps, and the station 2’s throughput falls to between 43 and 50 Mbps.The stations 3 and 4’s throughput varies from 45 to 65 Mbps. The overall adaptation is completed within 0.2 second.

These testing results indicate that the resource allocation scheme in Wi-fiis effective.

VI. CONCLUSION

In this paper, a software defined open Wi-fiplatform is designed and developed. Sphere decoding and Radix-4 Viterbi decoding are employed to enhance the system’s performance and efficiency. An interface is proposed and developed for signaling interaction between Wi-fiAP and 5G core network in cellular core network to support future network convergence. This open Wi-fiplatform supports fast prototyping and verification of new physical layer algorithms as well as the 5G+Wi-finetwork architecture and upper layer evolution.

ACKNOWLEDGEMENTS

This work was supported in part by the National Natural Science Foundation of China under Grant No. 61671436 and the Science and Technology Commission Foundation of Shanghai under Grant No. 15511102602,16511104204.

[1] D. J. Deng, S. Y. Lien, J. Lee, K. C. Chen, “On Quality-of-Service Provisioning in IEEE 802.11ax WLANs”,IEEE Access, vol.4, pp.6086-6104, April,2016.

[2] Microsoft Research Software Radio (Sora),https://www.microsoft.com/en-us/research/project/microsoft-=research-software-radio-sora/, January, 2017.

[3] LEI Tao, WEN Xiangming, LU Zhaoming, ZHAO Xing, LI Yangchun, ZHANG Biao, “SWN: An SDN Based Framework for Carrier Grade Wi-fiNetworks”, [J]China Communications, vol.13, no.3,pp.12-26, 2016.

[4] WARP: Wireless Open Access Research Platform, http://warpproject.org/trac, January, 2017.

[5] K. Kousias, K. Katsalis, D. Stavropoulous, T. Korakis, L. Tassiulas, “Building Virtual 802.11 Testbeds Towards Open 5G Experimentation”,IEEE Wireless Communications and Networking Conference(WCNC 2016), Doha, Qatar:IEEE Press,2016, pp.1-7.

[6] Open5GCore — The Next Mobile Core Network Testbed Platform, http://www.open5gcore.org/,January, 2017.

[7] 802.11-2016, “IEEE Standard for Information technology Telecommunications and information exchange between systems Local and metropolitan area networks, Speci fic requirements,Part 11: Wireless LAN Medium Access Control(MAC) and Physical Layer (PHY) Speci fications”,USA,IEEE, 2016.

[8] HUANG Zhengyong, XIA Jun, YU Hui, GUAN Yunfeng, GAN Xiaoying, LIU Jing, "Fusing Fixed and Hint Landmarks on Crowd Paths for Automatically Constructing Wi-fiFingerprint Database”, [J] China Communications, vol.12, no.1,pp.11-24, 2015.

[9] R. Zi, X. Ge, T. John, C.-X. Wang, H. Wang and T. Han, “Energy Eきciency Optimization of Frequency Radio Chains for 5G Wireless Communication Systems”,IEEE Journal Selected Areas Communications, vol.34, no.4, pp.758-771,2016.

[10] Wei H., Wang D., Wang. J. and You X., “Impact of RF mismatches on the performance of massive MIMO systems with ZF precoding”,Science China Information Sciences,vol.59, no.2, pp.1-14, 2016.

[11] T. Chiueh, P. Tsai, I-W Lai, “Baseband Receiver Design for Wireless MIMO-OFDM Communications”, USA,Wiley-IEEE Press, pp.209-249, 2012.

[12] B. Hassibi, H. Vikalo, “On the Sphere-decoding Algorithm I. Expected Complexity”,IEEE Transactions on Signal Processing, vol.53, no.8,pp.2806-2818, 2005.

[13] H. Vikalo, B. Hassibi, “On the Sphere-Decoding Algorithm II. Generalizations, Second-Order Statistics, and Applications to Communications”,IEEE Transactions on Signal Processing, vol.53,no.8, pp.2819-2834, 2005.

[14] A. J. Viterbi, “Convolutional Codes and Their Performance in Communication Systems”,IEEE Transaction on Communications, vol.19, no.5,pp.751-772, 1971.

[15] C.-H. Lin, C.-S. Yu, “Multimode Radix-4 SISO Kernel Design for Turbo/LDPC Decoding”,IEEE Transactions on Very Large Scale Integration(VLSI) Systems, vol.23, no.10, pp.2256-2267,2015.

[16] T. PEI, Y. DENG, Z. LI, G. ZHU, G. PAN, Y. CHOI,Hiroo SEKIYA, “A throughput aware with collision-free MAC for wireless LANs”,Science China(Information Sciences), pp.240-242, February,2016.

[17] M. Jo, T. Maksymyuk, R. L. Batista, T. F. Maciel,A. L. F. de Almeida, M. Klymash, “A survey of converging solutions for heterogeneous mobile networks”,IEEE Wireless Communications,vol.21, no.6, pp.54-62, 2014.

[18] M. A. Ridwan, N. A. M. Radzi, F. Abdullah, N. M.Din, M. H. Al-Mansoori, “Fiber Wireless testbed using Universal Software Radio Peripheral(USRP)”,2016 IEEE Region 10 Conference (TENCON), Singapore, IEEE, pp.3601-3604, 2016.

[19] USRP X310, https://www.ettus.com/product/details/X310-KIT, January, 2017.

[20] UBX 10-6000 MHz Rx/Tx (160 MHz, X Series only), https://www.ettus.com/product/details/UBX160, January, 2017.

[21] Medbo J, Schramm P, “Channel models for HIPERLAN/2 in different indoor scenarios”, Braneri085b, 1998.

[22] C.-X. Wang, S. Wu, L. Bai and etc. “Recent advances and future challenges for massive MIMO channel measurements and models”,Science China (Information Sciences), vol.59, no.2, pp.6-21, 2016.

[23] Lu X., Hui P., P. Lio, “Offloading mobile data from cellular networks through peer-to-peer Wificommunication: A subscribe-and-send architecture”,China Communications, vol.10, no.6,pp.35-46, 2013.

[24] X. Ge, T. Han, Y. Zhang, G. Mao, C-X. Wang, J.Zhang, B. Yang and S. Pan, “Spectrum and Energy Eきciency Evaluation of Two-Tier Femtocell Networks with Partially Open Channels”,IEEE Transactions on Vehicular Technology, vol.63,no.3, pp. 1306-1319, 2014.