A 10-Gbit/s Wireless Communication Link Using 16-QAM Modulation in 140-GHz Band

Transcription

1 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO. 7, JULY A 10-Gbit/s Wireless Communication Link Using 16-QAM Modulation in 140-GHz Band Cheng Wang, Changxing Lin, Qi Chen, Bin Lu, Xianjin Deng, and Jian Zhang Abstract This paper describes a 140-GHz wireless link whose maximum transmission data rate is 10 Gbit/s. A sub-harmonic mixer and multiplier based on Schottky barrier diodes, a waveguide H ladder bandpass filter, a Cassegrain antenna, and other components have been developed to construct a high-performance transmitting and receiving front end. 16 quadrature amplitude modulation has been adopted to improve the spectrum efficiency to 2.86-bit/s/Hz. A 32-way parallel demodulation architecture based on frequency-domain implementation of the matched filter and timing phase correction is proposed. An adaptive blind equalization algorithm is also realized to enhance the tolerance for channel distortion. The modulated signal is centered at GHz with 5-dBm output power. This link succeeded in transmission of a 10-Gbit/s signal over a 1.5-km distance with a bit error rate of 1e-6 in non-real-time mode. The measured 99.99% power bandwidth of the 10-Gbit/s signal is 3.6 GHz. The lowest acceptable signal noiserateperbit( ) is 15 db. This link also transmitted a 2-Gbit/s real-time signal with lowest. Index Terms Communication systems, millimeter-wave communication, quadrature amplitude modulation (QAM). I. INTRODUCTION B ROADBAND high-speed wireless communications have been widely used in mobile telephone, wireless local area networks (WLANs), satellite communication, and other consumer electronic equipments. With the development of fiber-optic techniques, the telecommunication companies can provide multi-gigabit/second data-transfer services for every family today. The demands for high-speed wireless access are increasing rapidly over recent years. Millimeter-wave and terahertz range from 100 GHz to 10 THz has sufficient bandwidth, which may be the location of future Gbit/s wireless communication systems. Wireless HD and IEEE c standards have been established in 60-GHz band in recent years. An interest group of terahertz (IG-THz) has also been organized under the IEEE wireless personal area network (WPAN) workgroup. The primal prototype of terahertz communication is realized through analog modulation. A 300-GHz analog video has Manuscript received January 30, 2013; revised April 28, 2013; accepted May 01, Date of publication May 31, 2013; date of current version June 28, This work was supported by the Terahertz Research Center, Institute of Electronic Engineering (IEE), China Academy of Engineering Physics (CAEP). The authors are with the Terahertz Research Center, Institute of Electronic Engineering (IEE), China Academy of Engineering Physics (CAEP), Mianyang , China ( c-w04@163.com). Color versions of one or more of the figures in this paper are available online at Digital Object Identifier /TMTT been transmitted over 22 m by Jastrow et al. [1]. In today s communication systems up to tens of gigabits/second, the modulation schemes are usually simple such as amplitude shift keying (ASK), on off keying (OOK), or binary phase-shift keying (BPSK) due to the high speed, simplicity, and reliability; however, they lead to low spectrum efficiency ( 1 bit/s/hz). A typical example is the 120-GHz 10-Gbit/s wireless link developed by NTT Microsystem Integration Laboratories [2], which adopts the ASK scheme. The maximum transmitted distance is 5.8 km [3]. The relative bandwidth is 17 GHz with spectrum efficiency of 0.59 bit/s/hz. A 135-GHz communication with 10-Gbit/s ASK modulation resembles the work of NTT in efficiency [4]. Progress of monolithic microwave integrated circuit (MMIC) technology could further promote the data rate of the direct modulator. The Fraunhofer Institute for Applied Solid-State Physics (IAF) has developed 220-GHz transmitter and receiver MMICs based on a 50-nm InP metamorphic high-electron mobility transistor (mhemt) [5]. The 10-m transmissions of the 25-Gbit/s OOK signal has been realized [6]. Communication based on uni-traveling carrier photodiodes (UTC-PDs) is another candidate in further terahertz communication, actually, a 300-GHz 24-Gbit/s data transmission link over 50 cm has also been realized by NTT, which also adopts ASK modulation [7]. Besides ASK and OOK, duobinary baseband modulation has also be implemented by Bell Laboratories, who transmitted a 2.5-Gbit/s data signal in 625 GHz with efficiency of 0.83 bit/s/hz [8]. Although millimeter-wave/terahertz range has sufficient bandwidth to accommodate wideband signal, high order modulation still has conspicuous advantages. Firstly, only vacuum traveling-wave tubes (TWTs) cascaded with solid-state power amplifiers can amplify modulated signal to tens of watts over 100 GHz [9] [12]. As the operation bandwidth of TWTs in the terahertz band is limited to several gigahertz, e.g., 5 GHz in the 220-GHz band [12], high-order modulation can efficiently increase the available data rate in a limited channel. High-order modulation is essential for high-speed far-distance millimeter-wave/terahertz communication. Secondly, compared with direct or simple modulation schemes, digital signal-processing techniques like pre-distortion and channel equalization could be used in high-order modulation, which help to enhance the system tolerance. Thirdly, it brings more flexibility in channel assignment for future application. However, high-order modulation is still a challenge for high-speed communication due to its complexity and channel distortion. A GHz 6-Gbit/s wireless link with an 8 phase-shift keying (8PSK) scheme and spectrum efficiency of 2.4 bit/s/hz has been developed by CSIRO, Highett, VIC, Australia [13] /$ IEEE

2 2738 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO. 7, JULY 2013 Free-space transmission of a millimeter-wave/terahertz wave suffers from high atmosphere attenuation. Research on sea-level clear atmosphere shows 1-dB/km attenuation in a 140-GHz atmosphere window, 2 3 db/km in a 220-GHz band, and 10 db/km in a 340-GHz band under air pressure of kpa, temperature of 15 C, and water vapor density of 7.5 g/m [14]. Higher transmission frequency will introduce higher attenuation. Fog and rain will also increase the possibility of communication interruption. Depending onthe output power of devices ( 30-W continuous-wave power is available by vacuum TWTs) and atmosphere attenuation, a 140-GHz band is considered to be a suitable location for high-speed far-distance communication like satellite to earth, inter-satellite, airborne, and deep-space link. This paper presents a 140-GHz wireless link, which transmitted a 10-Gbit/s signal over 1.5 km with. 140-GHz front ends have been developed, which consist of a high-performance sub-harmonic mixer (SHM), bandpass filter (BPF), high gain antenna, and local oscillator (LO) multiplier. A 16-QAM modulation scheme adopted in the current system could reduce the bandwidth of the 10-Gbit/s signal to 3.6 GHz, which is only 21% of NTT s 120-GHz 10-Gbit/s link ( GHz). Due to the complexity and limitation of high-speed digital logic circuits, the 10-Gbit/s signal has been demodulated by software in the non-real-time method firstly to validate the channel characteristics and processing algorithm. A 2-Gbit/s hardware real-time parallel demodulator has also been developed to validate the hardware realization of 32-way parallel demodulation. The 2-Gbit/s signal at 140 GHz has also been transmitted over 1.5 km. The measured minimum bit error rate (BER) is 1.80e-11 with RS(204, 188) coding. The experimental data indicate the lowest acceptable signal noise rate per bit ( )for 16 quadrature amplitude modulation (16-QAM) transmission is 15 db. Fig. 1. Schematic of 140-GHz wireless link Tx. II. SYSTEM ARCHITECTURE OF WIRELESS LINK Fig. 2. Schematic of 140-GHz wireless link Rx. The simplified block diagram of the 140-GHz link is shown in Figs. 1 and 2. This link can be described as two parts: the 140-GHz front end and signal processing platform. The 140-GHz front end is used to generate, transmit, and detect the carrier wave under given a 7.5-GHz IF signal. The available bandwidth of front end is 5 GHz. The signal processing platform is used to perform a modulation and demodulation algorithm under a given binary data sequence. The front end and the signal-processing platform are interfaced in an 7.5-GHz IF signal through a high-speed analog digital converter (ADC) and digital analog converter (DAC) with 10-bit resolution. Transmitter with high linearity, low noise receiver, and high directional antenna are developed to enhance detected ; parallel algorithms as timing/carrier synchronization and channel equalization are proposed to overcome the signal distortion. Multi-stage superheterodyne frequency mixing is adopted in a 140-GHz front end. The modulated 7.5-GHz IF carrier wave will be upconverted to GHz by an SHM. A -band doubler provides a 66.4-GHz LO signal for the mixer, which is driven by a 33.2-GHz dielectric resonance oscillator (DRO). The mixer and doubler are based on Schottky barrier diodes (SBDs). The LO frequency is one-half of the output carrier wave using sub-harmonic mixing. It brings advantages such as inter-modulation (IM) components reduction, fundamental and even harmonic components suppression, and more compact volume. The 140-GHz carrier wave is then filtered by a high sideband rejection H ladder BPF and then amplified to 5dBm by a MMIC amplifier. The output signal will suffer modest distortion, as 5 dbm is the 1-dB compression point of the transmitter. After being transmitted and received by high directional cassegrain antennas, the carrier will be down converted to 7.5 GHz again by the receiver before being delivered to the demodulation platform. High-speed ASK, BPSK, and QPSK modulation schemes could be achieved by direct modulation using analog devices. However, a high-order 16-QAM scheme can only be realized by digital signal-processing techniques. The relationship between modulation order and bandwidth is as follows: (1)

3 WANG et al.: 10-Gbit/s WIRELESS COMMUNICATION LINK USING 16-QAM MODULATION IN 140-GHz BAND 2739 TABLE I CHARACTERISTICS OF COMPONENTS IN 140-GHz TRANSCEIVER Fig. 3. Photograph of 140-GHz receiver module. is the data rate, is the rolloff factor, and is the modulation order. Rolloff factor is set to 0.4 in the current system. Therefore, the calculated bandwidth of 10-Gbit/s signal is 3.5 GHz. The relative spectrum efficiency is 2.86 bit/s/hz. According to theory of Nyquist, a double sample rate of highest operation frequency is required for recovering the carrier wave without distortion. The 10-Gbit/s signal in this link is processed by a software platform in non-real time mode. The primary binary data sequence is processed by RS(204, 188) coding and a 16-QAM modulation algorithm. The 7.5-GHz IF carrier wave with 3.5-GHz bandwidth is then directly sampled out by a high-speed DAC before being delivered to the transmitter. The sample rate of the DAC is 20 Gsample/s with a resolution of 10 bit, which is nearly two times higher than the IF carrier wave ( GHz). The received IF signal from a 140-GHz receiving front end will also be sampled by the ADC at the same rate, and then demodulated and decoded by the software demodulator. Data sequence of 2 Mbit will finally be recovered at each sample period. A 2-Gbit/s hardware demodulator is also developed besides the 10-Gbit/s software demodulator, which works in the real-time mode. The calculated bandwidth of the 2-Gbit/s signal is 700 MHz. Thus, it adopts a second stage mixing stage to further convert the IF signal from 7.5 to 0.5 GHz, which could be sampled by 2 Gsample/s on a broad ADC. The 10-Gbit/s signal is located at GHz. The 2-Gbit/s signal is located at GHz. III. 140-GHz FRONT-END PROTOTYPE A. Prototype Specification A prototype of a 140-GHz wireless front-end has been developed based on Schottky diodes and MMICs. The 140-GHz receiver is shown in Fig. 3. The characteristics of components developed for the 140-GHz transmitter and receiver is shown in Table I. The received RF signal will be amplified firstly by a low-noise amplifier (LNA) based on 0.1- m InP high-electron mobility transistor (HEMT) MMICs. A -band full-band isolator is inserted to get a better standing wave. Under the pumping of a 70-GHz doubler, the SHM converts the 140-GHz carrier wave to detectable IF frequency. The -band tunable attenuator is used to adjust the power of the LO signal to get optimal performance of conversion gain and noise performance. The measured output power of the 140-GHz transmitter is 5 dbm in the 1-dB compression point, as shown in Fig. 4 (left). The -parameter of the LNA measured by -band vector network analyzer (VNA) extension shows a concave point within Fig. 4. Measured output power of transmitter module (left); measured small conversion gain of transmitter/receiver (right) GHz. As a result, the variation of output power is 1.74 db in band. The variation of output power is measured to be 1.1 db. Actually, the LNA in the transmitter and receiver applies the same components. The power amplifier in 140 GHz is still under development, which may increase the output power to 10 dbm and extend the communication distance to about 6 km. The conversion gain of the transmitter is 2dBinaverage from to GHz. The noise temperature receiver is 1953 K through hot cold measurement with black body radiation sources. The conversion gain of the receiver is 0.5 db. An IF processing module is cascaded behind the receiver. It consists of a 50-dB high gain variable amplifier chain and second stage mixing module. A power supplier and controller are also integrated in the front-end. B. SHM and Doubler Based on SBDs A -bandshmbasedonantiparallelschottkydiodeshas been developed [15]. The mixer is designed as a microstrip circuit on a quartz substrate. The discrete SBDs are bonded to the circuits though silver epoxy. The parasitic components above 100 GHz of SBDs are difficult to be exactly modeled through lumped circuits. Thus, a new modeling method of SBDs based on a 3-D electromagnetic (EM) periphery structure model and lumped junction model is proposed, as shown in Fig. 5, the total structure of SBDs is established using the High Frequency

4 2740 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO. 7, JULY 2013 Fig D EM model of antiparallel SBD embedded in SHM circuit. Fig GHz doubler (left) and measured output power. Fig GHz SHM (left) [15] and measured SSB conversion loss. Structure Simulator (HFSS) software with an approximate parameter. A coaxial probe replaces the Schottky junction, which converts the 3-D EM SBDs model to a linear system. The 3-D EM model could be solved by an FEM simulator and indicates the variation of parasitic components exactly. The lumped junction model is established based on dc and ac measurement, which gives the nonlinear of the Schottky junction. A completed model of SBDs could be obtained by combining the two parts together in a harmonic-balance simulator. It coincides well in simulation and experiment. The available bandwidth of the 140-GHz SHM has been expanded to the entire -band ( GHz). Fig. 6 shows the measured and simulated single-sideband (SSB) conversion loss of 140-GHz SHM. The LO signal is set to 70 GHz with power of 3 4 mw. The measured lowest conversion loss is 7.9 db in the upper sideband and 7.1 db in the lower sideband. The 3-dB IF bandwidth is 22 GHz. The difference between simulation and measurement is lower than 0.9 db, which coincides well with each other. A 70-GHz doubler based on serial resistive SBDs has been developed, as shown in Fig. 7. Four diodes are cascaded to tolerate 200-mW driven power. The maximum measured output power is 27.8 mw in 67 GHz. The associated conversion efficiency is 13.9%. The output power in the GHz band is above21.5mwwithefficiency better than 10.8%. The phase noise of the 33.2-GHz DRO in the transmitter is kHz offset. The measured phase noise of the 7.5-GHz IF in receiver is kHz offset. It shows very low phase-noise addition by the 70-GHz doubler. The modeling method in the SHM design is also adopted in the design of the doubler. The difference between simulation and measurement is less than 0.5 db. C. Sideband Rejection BPF A 140-GHz waveguide H ladder BPF has been developed [16], as shown in Fig. 8. It is used to reject the lower sideband components of the SHM from to GHz and define Fig GHz H ladder waveguide BPF [16]. Fig GHz antenna housing (left) and Cassegrain antenna (right). the channel of this link. A seven-pole Chebyshev filter matrix synthesize technique is adopted in the design of the BPF, which introduces a seven half-wavelength resonator in the TE10 mode and eight coupling windows with a length of 0.5 mm. A modematching program is developed to simulate the -parameter of filter. As 5- m precision and 0.3- m surface roughness could be realized based on high-speed computerized numerical control (CNC) milling fabrication, extreme low insertion loss could be achieved. Fig. 10 (left)showsthe -parameter of this BPF. The simulated center frequency of the BPF is GHz, the 3-dB bandwidth is 11.8 GHz, and the passband return loss is better than 18.1 db. The measured center frequency is GHz. The 3-dB bandwidth is 11.6 GHz, and the relative bandwidth is 8.3%. The passband return loss is better than 16.4 db. The sideband rejection is better than 48.1 db below 130 GHz. They coincide well with each other, even the seven poles of return loss can be easily observed. The measured in-band insertion loss is 1.0 db, which is the lowest data in the literature. D. High-Gain Cassegrain Antenna A 140-GHz Cassegrain antenna and antenna housing have been developed, as shown in Fig. 9. The feed horn antenna is a wideband rectangular waveguide horn with 20-dBi gain.

5 WANG et al.: 10-Gbit/s WIRELESS COMMUNICATION LINK USING 16-QAM MODULATION IN 140-GHz BAND 2741 Fig. 10. Measured -parameter of 140-GHz BPF (left) and measured pattern of 140-GHz Cassegrain Antenna (right). Fig. 11. Architecture of the high-speed parallel demodulator. Although cross-polarization has been applied to accommodate the duplex signal in one antenna [2], a Cassegrain antenna in this system operates in the co-polarization mode. The aperture of this horn is 10.2 mm 7 mm. The diameter of the accessory reflector is 36 mm. The diameter of the main paraboloid is 360 mm. Fig. 10 (right) shows the pattern of this antenna. The measured far-field gain is 51.0 dbi. The 3-dB beam width is 0.6 in the horizontal direction and 0.35 in the vertical direction. The rejection between the main radiation lobe and secondary lobe is 9.4 db. The antenna housing is designed to isolate dust and humidity. It is made of polyethylene with 99.99% purity. The shape of the antenna housing is optimized to avoid distorting the pattern of the antenna. The surface of the antenna housing is plated with low radiation absorption material. The measured transmission loss of this antenna housing is only 0.2 db. The distortion of the antenna pattern introduced by the housing is inconspicuous by comparing the measured patterns of the antenna with housing and that with housing removed. The antenna is assembled on a tripod, which could provide 0.01 pinpoint precision. IV. PARALLEL DEMODULATION ALGORITHM AND IMPLEMENTATION A. 32-Way Parallel Demodulation Algorithm The requested for a given BER increases as modulation order. The influence of dust, water vapor, molecule absorption, and atmosphere will lead to certain free-space channel distortion on the transmitted signal. The phase noise of the LO, power amplifier linearity, group delay of the filter, and thirdorder inter-modulation (IM3) of the transmitter and receiver will introduce unpredictable distortion too. The in-band amplitude and phase distortion has a remarkable influence in high-speed communication. The 10-Gbit/s signal in the current system is modulated on a single carrier, not using multi-channel methods like CSIRO s link [13]. As a result, channel equalization is indispensable. Other algorithms, such as timing synchronization and carrier synchronization, are very important in high-order demodulation. Due to the limitation of field programmable gate array (FPGA) operation clock rate, the serial demodulation technique, which occupies huge calculation resources, is not suited to high-speed demodulation. We have proposed a new 32-way parallel architecture of demodulator based on the frequency-domain implement of matched filter (MF) and timing phase correction (TPC) in [18], which is shown in Fig. 11. The symbol rate of the 10-Gbit/s IF signal is 2.5 Gsymbol/s. In software demodulation, the IF signal is sampled by the ADC in 20 Gsample/s with 10-bit resolution. Sampled data is parallelized to 32 paths, which reduce the data rate to 625 Msample/s per path. These 32-path data are written to 32 parallel first-in first-outs (FIFOs), and the index controlled read operation of the FIFOs for timing frequency offset correction is made. Mixer-free downconverter and overlap operation are carried out in the next step before 64-point digital Fourier transition (DFT). 64-point DFT data are multiplied by DFT of the 33-point square root raised cosine (SRRC) MF, and then multiplied by a rotator to make TPC. 64-point inverse digital Fourier transition (IDFT) is then performed. The middle 32-point outputs of IDFT are transferred to operation and maintenance (O&M) timing error detector (TED). In digital communication, the demodulator must sample the signal at the maximum amplitude point of a symbol to enhance the signal-to-noise ratio (SNR). Therefore, a timing synchronization algorithm [19] is used to find the exact time point of the maximum amplitude point. The timing synchronization algorithm in the current system applies a dual feedback structure. After the timing error is calculated by O&M TED, the timing phase offset is fed back to realize TPC, and the timing frequency offset is fed back to control the reading index of FIFOs based on delete keep algorithm. Channel equalization is used to overcome channel amplitude and phase unbalance. The parallel adaptive blind equalization module PCMA will process the eight parallel symbols from O&M in the next step. This module is based on the relaxed look-ahead pipelined parallel adaptive constant mode analog (CMA) equalization algorithm. We have designed a new phase and frequency detector (PFD) based parallel decision feedback phase-locked loop (DFPLL) structure for carrier synchronization. Eight parallel symbols aresentto the DFPLL for carrier recovery. Finally, I/Q outputs are sampled. The demodulated binary data are decided in 16-QAM constellations. RS(204, 188) decoding is finally performed. Compared with demodulation, modulation of the 16-QAM signal is easier, which could be carried out in a more conventional way. Thus, the detail of modulation will not been included here. B. Implementation of Hardware Demodulator The implementation of the 10-Gbit/s demodulator depends on the sample rate of the ADC and DAC and the calculative capability of FPGAs. A 4 sample rate related to the symbol rate is required, which means a 10-Gsample/s ADC and DAC with 10-bit resolution is the prerequisite of the 10-Gbit/s demodulator (the symbol rate is 2.5 Gsymbol/s). The capability of

6 2742 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO. 7, JULY 2013 Fig. 12. Architecture and photograph of 16-QAM 2-Gbit/s demodulator broad. Fig. 14. Photograph of 140-GHz wireless communication link. TABLE II PERFORMANCE OF 140-GHz WIRELESS LINK Fig. 13. Simulated performance of 16-QAM demodulation algorithm and measured performance of the 2-Gbit/s parallel demodulator. FPGAs could be multiplied by the parallel FPGA array. Actually, a 10-Gbit/s ADC has already been developed by the Institute of Electronic Engineering (IEE), China Academy of Engineering Physics (CAEP), Mianyang, China. A 10-Gbit/s demodulator is feasible based on the current digital device. However, some challenges still remain, such as the peripheral interface, FPGA array, and reliability to be solved. The commercial available on-board ADC by the author could reach 10-bit resolution and a 5-Gsample/s ADC rate. Thus, a 2-Gbit/s hardware demodulator has been developed to validate the proposed algorithms with reduced data rate before a 10-Gbit/s demodulator. The architecture and photograph of the 16-QAM 2-Gbit/s demodulator is shown in Fig. 12, which consists of one ADC10D1500 chip and two V6LX240TFF1156 FPGAs. A PCI-e high-speed interface is integrated on broad to transfer binary data between the demodulator and personal computer. The sample clock of the ADC is 1 GHz, and the symbol rate is 500 Msymbol/s. The AD sample rate is 2 Gsample/s. Fig. 13 shows the simulated performance of the 16-QAM demodulation algorithm without RS(204, 188) coding. As the signal noise rate per bit ( ) reaches 14.3 db, the BER will be better than 1e-6. Measured demodulation performance of the 2-Gbit/s broad is also shown in Fig. 14, which requires 2 db more than simulation. V. MEASURED LINK PERFORMANCE under given BER A. 10-Gbit/s Non-Real-Time Transmission Fig. 14 shows aphotograph of the 140-GHz wireless link. The link performance is listed in Table II. The loss of the 1.5-km light-of-sight (LOS) channel is db based on (2) as follows: is the speed of light. The received power is 45.8 dbm calculated through (3) as follows: is the transmitter output power in dbm; and are the antenna gain in the transmitter and receiver; is the loss of the antenna feed line, which is 1.2 db in both the transmitter and receiver; and is the atmosphere attenuation, which is about (2) (3)

7 WANG et al.: 10-Gbit/s WIRELESS COMMUNICATION LINK USING 16-QAM MODULATION IN 140-GHz BAND 2743 Fig. 16. Measured constellation of 16-QAM 10-Gbit/s signal after 1.5-km transmission in 140-GHz band. Fig. 15. Measured of 10-Gbit/s signal in bench test and 1.5-km transmission test under given received RF power. 1dB/kmin140GHz.The after 1.5-km transmission is calculated as 19.9 db by (4) as follows: (4) is the Boltzmann constant: J/K. The received RF power is measured as 44.3 dbm after 1.5-km transmission, as shown in Fig. 15, which is 1.5 db higher than calculation. This is caused by uncertainties of instruments. Generally speaking, the measurement and theory prediction coincide well with each other. The atmosphere attenuation is not considered to have significant influence at this distance. 10-Gbit/s transmissions are performed to validate the channel distortion and demodulation algorithm. It should be noticed that a data frame of 2 Mbits is recovered at each time of non-real-time demodulation. The measured BER is considered to be 1e-6 when error bits have not been detected. The RS(204, 188) code has been performed in 10-Gbit/s transmissions, which has 92.2% coding efficiency. In theory, BER below 1e-10 could be achieved while the random BER is below 1e-3 and RS(204, 188) code is applied. The 140-GHz link was measured in the laboratory before 1.5-km far-field tests. A calibrated tunable attenuator is inserted between the transmitter and receiver to simulate the channel loss. The measured 99.99% power bandwidth of 10-Gbit/s signal is 3.6 GHz. As shown in Fig. 15, the detected is better than 16.4 db with the received RF power above 49.6 dbm in bench tests. Associated in 1.5-km distance is 18.9 db with 44.3-dBm received RF power. A deference of 2 db is presented between bench tests and far-field tests, which is also caused by uncertainties of measurement. Fig. 16 shows the demodulated constellations of the 10-Gbit/s signal after 1.5-km transmission. The transmitter operates in the 1-dB gain compression point. The constellations of timing synchronization, channel equalization, and carrier synchronization are shown, respectively. It could be noticed that the constellations are quite discrete before the equalizer, and the total phase of constellations is rotated in a certain angle before the carrier Fig. 17. Measured BER of 10-Gbit/s signal in bench test and 1.5-km transmission test under given. recovery. In the final step, the constellation concentrates at 16 amplitude and phase points. Gain compression of the transmitter could be found as the data points in four corners of constellations tend to move toward the center. These results could prove that the proposed algorithm is very efficient in correcting the channel distortion. The measured BER is in bench tests and the 1.5-km transmission experiment is shown in Fig. 17. In bench tests, the received RF is changed from 32.0 db (transmitter works in maximum output power) to 14.4 db using the inserted variable attenuator. The measured BER is better than 1e-6 (no error bits are detected in each 2-Mbit data frame) as db. As decreases to 14.4 db, the BER increases to 5.29e-5. In the 1.5-km far-field transmission experiment, the received RF power is attenuated by varying the IF input power toward the SHM. The measured data indicates that even when drops from 18.9 db (the transmitter works in maximum output power) to 14.5 db, no error bits have been detected. DuetothelimitedgainoftheIFamplifier in the receiver, no further data under lower have been obtained in both tests. db is very close to the threshold of errorfree demodulation. As a result, it is reasonable to be summarized from the bench test and far-field experiment that transmission with an acceptable BER can be achieved as db in 10-Gbit/s transmission.

8 2744 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO. 7, JULY 2013 Fig. 18. Measured of 2-Gbit/s signal in bench test and 1.5-km transmission test under given IF output power in GHz. Fig. 20. Measured BER of 2-Gbit/s signal in 1.5-km transmission test with Tx IF input power reduced. Fig. 19. Measured BER of 2-Gbit/s signal in bench test under given. B. 2-Gbit/s Real-Time Transmission The measured 99.99% power bandwidth of the 2-Gbit/s signal is 730 MHz. The measured of the 2-Gbit/s signal in bench tests and 1.5-km far-field tests are shown in Fig. 18. The measured is better than 15.6 db as the received RF power 56.6 dbm in bench tests. The measured is 23.5 db with 5-dBm RF transmitted power in the 1.5-km transmission experiment. It also shows a difference about 2 db compared with the bench tests. Fig. 19 shows the measured BER of the 2-Gbit/s signal in bench tests, which is performed by a hardware demodulator in a real-time method. It compares the data of the BER between that with or without RS(204, 188) decoding. The transmitter works in 5 dbm (1-dB compression point). The received is tuned by changing the attenuation of the variable attenuator. It needs to be noticed that the 1-dB compression point of the transmitter introduces more nonlinearity to the final signal. While the RS(204, 188) coding has not been included in 2-Gbit/s real-time demodulation, the minimum BER can only reach 1.6e-4 even in the highest. As the received is better than 15.6 db, the BER will be lower than 1.3e-3 in bench tests. It could be noticed that RS(204, 188) code significantly improves the BER. After RS(204, 188) decoding is adopted, as the is better than 12.9 db, the BER is better than 1.39e-7. The lowest BER is obtained to be 1.01e-10 with relative db. While db, the BER is between 1.99e-10 and 1.01e-10. Although the has already been sufficient, the recorded BER is dominated by the nonlinearity of the transmission channel. Fig. 20 shows the measured BER of 1.5-km transmission. The BER of 1.5-km transmission without decoding is 3.5e-4 at the output point. The received is tuned through reducing the IF input power to the SHM of transmitter. As a result, the RF output power, received, and nonlinearity of the signal will decrease synchronously with the reduction of the transmitter IF input. As the decreasing of and increasing of signal linearity, the BER will be further reduced to 1.10e-7 under db. The transmission does not apply RS(204, 188) coding. could be achieved under ultra-low, which has been measured to be 10.5 db. As the RS(204, 188) decoding is performed, the lowest BER reaches 1.80e-11 with db. While db, the BER is better than 2.11e-7. It should to be noticed that the BER increases to 1.1e-10 when the transmitter is driven gradually to the 1-dB compression point. At that moment, the nonlinearity of the transmitter dominates the BER. The BER in relative low ( 15 db) is dominated by Gaussian white noise. All in all, the theory and experimental results indicate that the lowest of 15 db is required for future applications. Table III gives the comparison between recently published millimeter-wave and terahertz communication links. It indicates that the highest spectrum efficiency of 2.86 bit/s/hz has been realized in current 140-GHz communication links. The efficiency of the simple modulation scheme is lower than 1 bit/s/hz. The 1.5-km transmission distance in this work is only shorter than the NTT 120-GHz 5.8-km link, but it needs to be indicated that the output power of this work ( 5dBm)is21dBlowerthan NTT 120-GHz link (16 dbm). As cascaded with solid-state and TWT amplifiers in the future, a communication link is possible

9 WANG et al.: 10-Gbit/s WIRELESS COMMUNICATION LINK USING 16-QAM MODULATION IN 140-GHz BAND 2745 TABLE III SUMMARY OF RECENTLY PUBLISHED MILLIMETER-WAVE/TERAHERTZ COMMUNICATION LINKS with an output power over 10 W. A 10-Gbit/s satellite to earth link may be realized in the near future. A communication link in 340 GHz has also been developed recently by CAEP, which also adopts 16-QAM modulation. A 340-GHz front end, 3-Gb/s hardware modulator, and demodulator have been realized. It has transmitted a 3-Gb/s signal over 100 m in the real time mode successfully. VI. CONCLUSION A 140-GHz wireless link with 16-QAM modulation has been described in this paper. A high-performance 140-GHz front end has been developed. 10-Gbit/s non-real time based on a software modulator and 2-Gbit/s real-time transmission based on a hardware modulator have been realized in bench tests and 1.5-km far-field tests. Under 5-dBm output power, the measured BER was 1e-6 for the 10-Gbit/s signal and 1.01e-10 for the 2-Gbit/s signal. The occupied bandwidth of the 10-Gbit/s signal is only 3.6 GHz. Its spectrum efficiency achieves 2.86 bit/s/hz, which is better than that of most current links. To the authors knowledge, this work is the first high-speed communication link above 100 GHz based on higher order modulation. Cascaded with solid-state power amplifiers and TWTs could further push the output power to level of tens of watts. The proposed link has the ability of extending the distance to over 100 km in the future. ACKNOWLEDGMENT The authors would like to thank C. Zhou, J. Yao, W. Su, S.Xiao,J.Liu,J.Yan,S.Wu,B.Cui,L.Miao,andJ.Jun,all with the IEE, CAEP, for their support. REFERENCES [1]C.Jastrow,K.Munter, R. Piesiewicz, T. Kurner, M. Koch, and T. Kleine-Ostmann, 300 GHz channel measurement and transmission system, in 33rd Int. Infrared, Millim. Terahertz Waves Conf., 2008, pp [2] A. Hirata, T. Kosugi, H. Takahashi, J. Takeuchi, H. Togo, M. Yaita, N. Kukutsu,K.Aihara,K.Murata,Y.Sato,T.Nagatsuma,andY.Kado, 120-GHz-band wireless link technologies for outdoor 10-Gbit/s data transmission, IEEE Trans. Microw. Theory Techn., vol. 60, no. 3, pp , Mar [3] A. Hirata, T. Kosugi, H. Takahashi, J. Takeuchi, K. Murata, N. Kukutsu, Y. Kado, S. Okabe, T. Ikeda, F. Suginosita, K. Shogen, H. Nishikawa, A. Irino, T. Nakayama, and N. Sudo, 5.8-km 10-Gbps data transmission over a 120-GHz-band wireless link, in 2010 IEEE Int. Wireless Inform. Technol. Syst. Conf., 2010, pp [4] Z. Bo, X. Yong-Zhong, W. Lei, and H. Sanming, A switch-based ASK modulator for 10 Gbps 135 GHz communication by 0.13 m MOSFET, IEEE Microw. Wireless Compon. Lett., vol. 22, no. 8, pp , Aug [5] I. Kallfass, J. Antes, T. Schneider, F. Kurz, D. Lopez-Diaz, S. Diebold, H. Massler, A. Leuther, and A. Tessmann, All active MMIC-based wireless communication at 220 GHz, IEEE Trans. Terahertz Sci. Technol., vol. 1, no. 2, pp , Nov [6] I. Kallfass, J. Antes, D. Lopez-Diaz, S. Wagner, A. Tessmann, and A. Leuther, Broadband active integrated circuits for terahertz communication, in 18th Eur. Wireless Conf., 2012, pp [7] H. J. Song, K. Ajito, Y. Muramoto, A. Wakatsuki, T. Nagatsuma, and N. Kukutsu, 24 Gbit/s data transmission in 300 GHz band for future terahertz communications, Electron. Lett., vol. 48, pp , [8] L. Moeller, J. Federici, and S. Ke, THz wireless communications: 2.5 Gb/s error-free transmission at 625 GHz using a narrow-bandwidth 1 mw THz source, in XXXth URSI Gen. Assemb. Sci. Symp., 2011, pp [9] L. A. Samoska, An overview of solid-state integrated circuit amplifiers in the submillimeter-wave and THz regime, IEEE Trans. Terahertz Sci. Technol., vol. 1, no. 1, pp. 9 24, Sep [10] V.Radisic,K.M.K.H.Leong,S.Sarkozy,X.Mei,W.Yoshida,L. Po-Hsin,andR.Lai, A75mW210 GHz power amplifier module, in IEEE Compound Semiconduct. Integr. Circuit Symp., 2011, pp [11] J. H. Booske, R. J. Dobbs, C. D. Joye, C. L. Kory, G. R. Neil, P. Gun-Sik, P. Jaehun, and R. J. Temkin, Vacuum electronic high power terahertz sources, IEEE Trans. Terahertz Sci. Technol., vol. 1, no. 1, pp , Sep [12] J.C.Tucek,M.A.Basten, D. A. Gallagher, and K. E. Kreischer, 220 GHz power amplifier development at Northrop Grumman, in IEEE 13th Int. Vacuum Electron. Conf., 2012, pp [13] V. Dyadyuk, J. D. Bunton, J. Pathikulangara, R. Kendall, O. Sevimli, L. Stokes, and D. A. Abbott, A multigigabit millimeter-wave communication system with improved spectral efficiency, IEEE Trans. Microw. Theory Techn., vol. 55, no. 12, pp , Dec [14] T.Schneider,A.Wiatrek,S.Preussler,M.Grigat,andR.Braun, Link budget analysis for terahertz fixed wireless links, IEEE Trans. Terahertz Sci. Technol., vol. 2, no. 2, pp , Mar [15] C. Wang, X. Deng, L. Miao, and J. Yan, GHz sub-harmonic mixer based on Schottky barrier diodes, in Int. Microw. Millim. Wave Technol. Conf., 2012, pp [16] C. Wang, B. Lu, J. Liu, and X. Deng, 140 GHz waveguide H ladder bandpass filter, in Int. Microw. Millim. Wave Technol. Conf., 2012, pp [17] C. Wang, L. Changxing, C. Qi, D. Xianjin, and Z. Jian, 0.14 THz high speed data communication over 1.5 kilometers, in 37th Int. Infrared, Millim., Terahertz Waves Conf., 2012, pp [18] C. Lin, B. Shao, and J. Zhang, A high data rate parallel demodulator suited to FPGA implementation, in 18th Int. Intell. Signal Process. Commun. Syst. Symp., Chengdu, China, Dec. 2010, pp [19] C. Lin, B. Shao, and J. Zhang, A multi-channel digital programmable delay trigger system with high accuracy and wide range, in Int. Electron., Commun., Control Conf., Ningbo, China, Sep. 9 11, 2011, pp

10 2746 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 61, NO. 7, JULY 2013 Cheng Wang was born in Suining, China, on March 8, He received the B.S. degree in engineering physics from Tsinghua University, Beijing, China, in 2008, and the M.S. degree in radiophysics from the China Academy of Engineering Physics (CAEP), Mianyang, China, in In 2011, he joined the Institute of Electronic Engineering (IEE), CAEP, Mianyang, China. His current research involves millimeter-wave/terahertz communication systems, solid-state mixers and multipliers based on Schottky diodes, and terahertz passive filters and circuits. Changxing Lin was born in Chongqing, China, on January 7, He received the B.S. degree in engineering physics from Tsinghua University, Beijing, China, in 2007, and the Ph.D. degree in nuclear science and technology from Tsinghua University, Beijing, China, in From June 2009 to December 2009, he was a Research Assistant with the European Organization for Nuclear Research (CERN). In 2012, he joined the Institute of Electronic Engineering (IEE), China Academy of Engineering Physics (CAEP), Mianyang, China. His current research involves algorithm and implementation of high-speed demodulation and terahertz wireless local area networks (WLANs). Qi Chen was born in Chongqing, China, on November 3, He received the B.S. degree in remote sensing techniques and instruments and M.S. degree in electromagnetic and microwave techniques from Xidian University, Xian, China, in 2003, and 2007 respectively, and is currently working toward the Ph.D. degree in radiophysics at the Institute of Electronic Engineering (IEE), China Academy of Engineering Physics (CAEP), Mianyang, China. In 2007, he joined the IEE, CAEP. His research interests include millimeter-wave/terahertz antennas, photo crystals, and metamaterials. Bin Lu was born in Chongqing, China, on August 18, He received the B.S. degree in electronic engineering from Fudan University, Shanghai, China, in 2004, and the M.S. degree in communication and information systems from the China Academy of EngineeringPhysics(CAEP),Mianyang,China,in2011. Since 2011, he has been an Engineer with the Institute of Electronic Engineering (IEE), CAEP, where he is engaged in research on terahertz communication and radar systems, millimeter-wave passive filters, and digital processing algorithms. Xianjin Deng was born in Anyue, China, on June 11, He received the B.S. degree in electronic engineering from Xidian University, Xian, China, in 1998, and the M.S. degree in electronic science and techniques from the University of Electronic Science and Technology of China, Chengdu, China, in In 2003, he joined the Institute of Electronic Engineering (IEE), China Academy of Engineering Physics (CAEP), Mianyang, China. His current research involves millimeter-wave/terahertz communication system, millimeter-wave solid-state power combining, and microwave active circuits. Jian Zhang was born in Sichuan, China, on November 27, He received the B.S. degree in electronic technology from the National University of Defense Technology, Changsha, China, in 1989, the M.S. degree in communication engineering from the China Academy of Engineering Physics (CAEP), Mianyang, China, in 1994, and the Ph.D. degree in electrical engineering from Chongqing University, Chongqing, China, in In 1989, he joined the Institute of Electronic Engineering (IEE), China Academy of Engineering Physics (CAEP), Mianyang, China. His current research involves electronic systems, wireless communication, and terahertz science and technology.