Design, implementation, and test of an SDR for NVIS communications

Although many physical layer solutions have appeared for remote sensors and Internet of things during the recent years, none of them is suited to very remote sensors in areas away from any mobile operator coverage. In that case, a solution on the basis of near vertical incidence skywave (NVIS) with reflection in the ionosphere may be very attractive. Using NVIS, no line of sight is needed and the coverage is much bigger than any other system operating in either the very high frequency (VHF) or ultra high frequency (UHF) band. In this paper, we present a new transmission scheme for very remote sensors using the NVIS transmission technique.

(NGOs) in terms of voice and e-mail communications, as well as e-health and educational initiatives. 6 Until not long ago, the antennas used for HF communications were large in size, especially for the lower part of the HF band (3)(4)(5)(6)(7)(8)(9)(10). There has been a significant effort in the research of compact HF antennas, with gains not lower than −20 dB that can be used for NVIS implementations, under the premise that a large antenna with a positive gain is placed at the other side of the link. [7][8][9] This fact opens the door for NVIS communications to be applied to remote sensing applications, since a sensor is usually a compact device, working in a low-rate low-power scenario. In that case, the concept of Internet of Things (IoT) can be extended to devices located in areas without any telecommunication infrastructure. Currently, the main solutions for IoT 10 are Sigfox, LoraWan, and NBIoT. Although the coverage of these networks is large (tens of kilometers), there is no global coverage in remote areas. In that case, satellite and HF communications are the only feasible solutions. Satellite solutions are usually too expensive and may not work under heavy weather conditions, such as snow and rain. On the other hand, HF communications operate at low, medium, and high bit rates (up to 62 kbps), with high transmission power (up to 250 W) using commercial equipment. 11,12 The concept of NVIS applied to remote sensors for remote IoT applications needs a new physical layer able to transmit low data rate with the minimum power consumption and compact and lossy antennas.
Most of the projects 13,14 carried out in the Antarctica need to install data loggers in remote places. These places have complicated access for scientists and it may take some days or even weeks to get this data. In the case of Antarctic stations that are only open during the Antarctic summer, some sensors should be operational the rest of the year and send their data to the station.
In this document, we propose the best modulation scheme for the application of NVIS communications for sensors located at a distance of up to 250 km from the data collection center without line of sight and low consumption platform and low bit rate (adapted for this type of applications). In particular, we propose a compact NVIS platform for RIoT that establishes an NVIS sensor network in Antarctica capable of sending information to the Spanish Antarctic Station. 15 The proposed platform based on SDR allows us to have flexibility in the development of the system to achieve low energy consumption.
Obviously, this can be also applied to any other unattended sensor placed in zones without mobile telephony coverage, as well as for the rescue of injured people in remote places. This extends the concept of Internet of Things (IoT) to very remote sensors. 16 This paper is organized as follows. In Section 1, a general description of the system is presented. In Section 3, a detailed description of the hardware implementation is given. The performed tests to select the proper power and modulation are described in Section 4, and the results are presented in Section 5. Finally, Section 6 contains the conclusions and future steps.

SYSTEM DESCRIPTION
Up to now, most of the sensors installed by the Spanish scientific community in Antarctica have been placed close to the stations, because that makes data collection and data transmission easier. The research project "Optimized HF transmission for Near Vertical Incidence Skywave (NVIS) links for remote sensors in Antarctica (ENVISERA)" 15 is a step forward for the communication capabilities of Spanish scientists in Antarctica. The main goal is the design of a sensors network able to transmit their data from very remote places (even in the continent) to the Antarctic Station Juan Carlos I. This network will use the NVIS that allows the communication between nodes up to 200 km away using vertical reflection in the ionosphere in the HF band . As both the transmitted and received signal come from the upper part of the atmosphere, no line-of-sight is needed and you can have any obstacle without any loss of the signal. Then, every single node can behave as a repeater if necessary, so large areas can be covered, provided that the distance between two contiguous nodes is less than 200 km. The transmitted power for each node is low, and that improves the autonomy of the system. The sensor network will also allow bit-rates up to tens of kbps, good enough for most of the sensors for the transmission of data and medium/high quality digital voice.
A sensors network like this makes the area of influence of the Antarctic stations wider, and makes new experiments no possible until now feasible. It also has straightforward applications in case of natural disasters and terrorism attacks, where HF networks using NVIS are the most agile and economic solution to communicate the affected area with other parts of the country.

DESCRIPTION OF THE TEST BED
The block diagram of the system can be seen in Figure 1 The frequency was selected after a detailed analysis of the ionograms from the Ebre Observatory in Roquetes, 17 situated 80 km from Cambrils. As referred in Lavers, 18 we selected the optimum transmission frequency as: where f oF2 is the average of the critical frequency of the F2 layer along the previous 15 days. The results showed in this paper were obtained at f TX = 5.4 MHz. In future developments, the sensor and the data collection center will implement some kind of automatic link establishment (ALE) 12 strategy in order to select the best frequency automatically. Two horizontal half-wave dipoles antennas were installed at both ends of the link between two 8 m high masts. The final height was adjusted through NEC simulation 19 to maximize the gain at 90 • vertical elevation. The final value was close to 10 , so around 6 m. At the receiver side, the signal is filtered and amplified 30 dB, with a low noise amplifier before the ADC. It should be noted that the presence of strong interfering signals come from the short wave stations, which can easily saturate the receiver if not properly filtered.
As the system is intended to be software-defined radio at low cost, the core of the transceiver is a Red Pitaya platform based on a Zynq SoC architecture of Xillinx. The Zynq-7000 All Programmable SoC (AP SoC) family integrates the software programmability of an ARM-based processor (the Processing System PS) with the hardware programmability of an FPGA (the programmable logic, PL).
The Red Pitaya platform 20 allows us to develop the hardware in a very agile way for different purposes. First, we programmed the platform to test the NVIS channel. After that, we tested the behavior of several modulations under different values of the transmission power. Finally, we implemented the optimal physical layer in real-time operation. The upper part named as FPGA is the PL, which takes care of all the fast digital signal processing. The lower part named SO is the processing system (PS), which includes the operating system, the configuration files, and the management of peripherals. The two entities share access to the RAM memory through the direct memory access (DMA) 24 device. The DMA is also in charge of the communication among the layers that exchange the data between the ADC and DAC converters.
A phase locked loop 23 (PLL) is used to synchronize all the different parts of the hardware design, while implementing the local oscillator for the up and down conversion of the radiofrequency signal. As the operating system (OS) and hardware (FPGA) share the same peripherals 24 (DDR RAM), an agile reconfiguration of the transceiver can be carried out. In particular, the center frequency, the transmission power, and the power range of the received signal are the most critical parameters that are changed dynamically. The USB port is used by the RP to control the GPS, the power amplifier, and the external memory via a USB hub. The GPS is used for time synchronization purposes, although is not strictly necessary, the power amplifier can be switched on and off and returns the current output forward and reverse power. The external memory is used for storing the received data for further analysis.

Up and down converters
The RP has a couple of analog inputs and digital outputs of 14 bits and 125 Msps that we will use for MIMO and diversity applications. While in the analog inputs there is one ADC per channel, there is only one DAC that has to be multiplexed and synchronized for the two outputs. The digital up converter (DUC) and the digital down converter (DDC) perform the conversion from a base band signal to an RF signal and vice versa. 25

FIGURE 2 System design
In order to improve the performance of the DDC and DUC, we need to test some possible configurations, changing the order of the CIC's filters and the oversampling factor 26 in each state. The alias rejection of the CIC filters 27 has to be coupled with the LSB of the ADC converters, in this case, −87 dBm. Finally, the values of the total filter response have to be below the alias rejection. In Table 1, we can see all the tested configuration values for the CIC filters, the bandwidth, and the alias rejection. The main part of the FPGA power consumption comes from the CICs stages. From Table 1, it follows that the R factor of each stage mainly affects the power consumption, while the alias rejection is dependent on the bandwidth. In the cases where the alias rejection response does not fall within the necessary margins for a good performance (less than −87 dB), the power consumption is not calculated. The block diagram of the complete designed system can be seen in Figure 2.
The selected configuration is shown in Figure 3. The first cascaded integrator comb (CIC) filter performs the down-sampling with a factor 50 from the original 125 Msps to 2.5 Msps. It is a third-order filter with a minimum alias rejection of 87 dB for a signal of bandwidth up to 100 kHz. The second CIC filter performs the down-sampling with a factor of 5 from 2.5 Msps to 500 ksps, with a fifth-order filter. Finally, the raised cosine filter (RCF) is an FIR filter of 79 coefficients, with a down-sampling factor of 5 and a bandwidth of 30 KHz, which compensates the distortion caused by the CIC filters in the pass-band and perform the down-sampling from 500 to 100 ksps. We decided to limit both the bandwidth and the slope of the filter in order to minimize the energy consumption. In those conditions, the average current consumption of the hardware is of 1.88 W and of the full ZYNQ system design is 0.85 A with a current peak of 1 A for a supply voltage of 5.00V (4.25 W).

RF amplifier
The class C amplifier can achieve high efficiency values (90%-100%) so it is commonly used in those applications that do not need a linear amplification. This kind of amplifier is particularly suitable for constant envelope modulations such as FSK or BPSK. However, in our work we wanted not only to improve the power efficiency but also to find the best modulation (FSK, PSK, and QAM) suited to the NVIS channel. For that reason, our platform had to be able to test both constant and not constant envelope modulations, so we needed a linear amplifier. Finally, we selected a Class A amplifier that allow us to perform all the tests shown in Table 2. The amplifier was a BLWA 0103-250, from BONN Elektronik GmbH, with a maximum transmission power of 250 W and a maximum efficiency of 30% (Alsina et al 21 ). The main features of the amplifier are described in 22

The application
The application has to control all the peripherals shown in Figure 1 for both transmission and reception. The GPS, the power amplifier and hard drive are external devices while the DMA and the shared registers are implemented in the PL part. As the transmissions are launched at regular intervals, the Linux tool Crontab is used to execute the application. The program reads the configuration file and updates the time reference from the GPS. Then it programs the FPGA to start transmission or reception, even if the buffers are empty. After that, the DMA is configured to move data from the RAM to the FIFO or vice versa and, in case of transmission, the power amplifier is activated. Finally, when all the data have been transmitted or received, the FPGA is blocked until the next program execution.
As we explain in Section 1, the power amplifier is a up to 250 W class A linear amplifier, able to test any proposed modulation. In case of using a constant envelop modulation, it could be replaced by a class C amplifier with a remarkable increase of the efficiency. This a key issue for unattended solar-powered sensors. We chose up to 100 W as an initial value to perform all the tests, but, as it will be explained later, the output power could be reduced significantly.

DESCRIPTION OF THE TESTS
Because of their very low power consumption, the battery of the remote sensors usually supplies little power. In order to not increase the requirements of the battery significantly, the transmission system using NVIS has to be power efficient.
The tests performed are designed to optimize the required power for bit rates lower than 4.65 kbps, which cover the major part of remote sensing applications. The tests compare the performance of simple modulations, ie, M-FSK and QAM, to find out the best modulation to be used with low transmission power. As you can see in Table 2, several modulations have been tested for different power levels ranging from 790 mW to 100 W. The tests start every 4 minutes, so 12 tests are sent every hour. The duration for each test is adjusted so as the same amount of bits is sent for each modulation. The occupied bandwidth is 2.3 kHz to be consistent with the most common HF standards. 28 The results presented in Section 5 are derived from a 2-month survey between October and December 2017. As we have only performed the transmissions during the day, the system only can use a single frequency.

RESULTS
In this section, we introduce the results from the measurement survey. First, the variation of the received power and the bit error rate (BER) as a function of the time of day. Second, the variation of BER as a function of the measured ratio between the bit energy E b and the noise spectral density N 0 . Finally, the statistics of the received BER are presented in terms of cumulative density function.

Time variation
The NVIS communication channel shows strong time variations depending on the time of the day and the season of the year. Although estimating the range of frequencies that may work well for a given time of the year is not easy, there is always a long time window, around 6 hours, during day and night where the link may be established without changing the frequency. In Figure 4A, you can see the average variation of the received power as a function of the time of the day and the transmitted power for a BPSK modulation. For every hour, the transmitted power ranges from 100 W (50 dBm) to 790 mW (29 dBm) as detailed in Table 2. For a fixed frequency, there are always 6 hours of channel availability between 10 AM to 16 PM. The received power ranges from −60 to −80 dBm, so a total loss of 105 dB is introduced by the channel. It is important to note that the received power increases at both the beginning and the end of the day, while it decreases around 1 PM.
In Figure 4B, you can see the average variation of the E b ∕N 0 as a function of the time of the day and the transmitted power for a BPSK modulation. For a transmitted power of 24 W, the values range between 25 and 30 dB, and for a transmitted power of 6 W, the values of E b ∕N 0 are around 20 dB most of the time. The E b ∕N 0 also decreases at certain hours, and increases at both the beginning and the end of the day.
During the night, the layer F2 is deionized, and the critical frequency decreases. If you want to transmit information during the night, you need to change the band and install a lower frequency antenna.

BER versus E b N 0
In this section, we evaluate the ratio bit energy E b over noise power spectral density (N 0 ). This ratio is measured in every transmission interval using (2d). The bandwidth has been limited to 3 kHz by a digital filter and the sampling frequency f m is 100 ksps.
As the NVIS channel shows multipath at some parts of the day and a high doppler spread as we can see in Orga et al, 29 the recursive least squares (RLS) equalizer has been adopted for all the modulation schemes, as suggested in Wang et al. 30 The results can bee seen in Figures 5A,B. For weak received signals, the performance of all modulations is similar. For a BER of 10 −1 , an E b ∕N 0 around 5, 6, and 8 dB is needed for BPSK, BFSK, and 4PSK, respectively.
In the range of E b ∕N 0 , between 8 and 12 dB, BPSK performs better than BFSK, with a difference of 4 dB for a given BER.
For higher values of E b ∕N 0 around 18 dB, BPSK and BFSK curves get closer at BER values around 10 −2 . We also have compared the measured performance with theoretical performance of these modulations in a Rayleigh channel scenario. The results can be seen in Figure 5B. The differences are because of the occurrence of strong fadings in the middle of the packet. As we can see during the transmission, the channel presents a mixture of Good, Moderate, and Poor conditions, as defined in Antoniou et al. 11 The measured BER values fit quite well with the theoretical values for the poor and moderate channel model.

CDF function
For a deeper analysis of the performance of phase versus frequency modulations independent of the transmitted power and the received E b ∕N 0 , the statistics of the BER have been evaluated.  Figure 6A,B, the cumulative density function (CDF) of the BER for BFSK and BPSK is evaluated as a function of the transmitted power. For BFSK, the probability of having a BER less than 10 −3 is higher than 0.81 for a transmitted power of 24 W. For a transmitted power of 6 W, this probability decreases to 0.6. For BPSK, the probability of having a BER less than 10 −3 is higher than 0.82 for a transmitted power of 24 W. For a transmitted power of 6 W, this probability decreases to 0.73 and the value is 0.5 for a transmitted power of 3 W.
We also note that 4FSK should perform better, since it concentrates all the power in a single subcarrier for any given time. However, the bandwidth needs to be larger so the orthogonality between subcarriers is preserved. For a given bandwidth and bit-rate, BPSK and BFSK perform much better.
Considering these results, we conclude that a transmitted power up to 6 W with BPSK modulation performed with the RLS equalizer is enough for remote sensors using NVIS communications. We have to consider that no error code correcting (ECC) has been used. If an ECC were used, the BER probability would approach to zero most of the time.

Power consumption
In order to optimize the power consumption, we have to choose the minimum transmission power to ensure a BER of 10 −2 , along a significant time window. From Figure 5B, we can state that an EbN0 greater than 16 dB is needed for a BPSK. For a transmission power of 6 W, we can achieve an EbN0 greater than 16 dB from 10 to 14 UTC, while the time interval is increased from 10 to 16 UTC if we transmit more than 12 W (see Figure 4B).
In the receiver front-end, the LNA is running most of the time so the optimization of its power consumption is a quite critical issue. As the received expected power is near to the LSB of the Red Pitaya ADC converters (around −85 dBm) for the worst case, we set a minimum fading margin of 24 dB to ensure a correct demodulation of the received signal. The LNA ZFL-500LN+ from mini-circuits 31 offers a gain of 24 + −0.5 dBs with a power consumption of 0.9 W. The power consumption of the FPGA, ARM, and the peripherals is 1.88 W, 2.4 W, and 1.72 W, respectively, making a total of 6 W. In reception mode, we have to add the consumption of the LNA, so the total is 7 W. In transmission mode, if we assume a C-class amplifier of 15 W with a 90% of efficiency, the consumption of the RF power amplifier will be 17 W. As the LNA is always running, we have to add the consumption of both amplifiers, so the total is 24 W.

CONCLUSIONS
In this paper, the feasibility of the application of NVIS communications to a remote sensing scenario has been investigated. Although the system is intended to be installed in remote sensors around the Spanish Antarctic Station, the system fits in any deployment with sensors located hundreds of km away from the hub without any available communication infrastructure. A new low-cost hardware using a commercial software radio platform has been implemented. We would like to note that, up to our knowledge, no similar developments have been made with the Red Pitaya platform in terms of complexity and communication speed between the PS and the PL.
The SDR approach has allowed us to use the same hardware platform for both channel sounding, modulation testing and the final implementation of the modem. The implementation of the system on a ZYNQ platform (FPGA + ARM processor) makes it possible to design accurately the CIC filters, peripherals management, and RF amplifier in a single hardware with a low power consumption.
On the basis of the study of the variation of the received power and EbN0, we have established a daily time window of about 6 hours for a fixed frequency. That is a key issue in order to optimize the antenna for narrowband, since the higher the antenna gain, the lower the transmitted power.
A test bed has been put into operation in a 100 km NVIS link in order to select the best modulation for low values of transmission power. BPSK modulation is the best choice because of its trade-off between BER performance and the efficiency of the power amplifier. We have determined that the use of BPSK with a transmission power less than 10 W guarantees the proper demodulation of the sensor data. As the BPSK is a constant envelope modulation, we can use a class-C amplifier with a power consumption lower than 15 W.
Our work proposes a design basis for scalable SDR platforms for very low power consumption, which can be applied not only for remote sensors, but also in emergencies, rescue of people, and communication in developing countries.
The following steps are the definition of a complete physical layer for remote sensing using NVIS communications. That includes the frame design, equalization, coding, and frequency selection. A new test bed will be installed around the Spanish Antarctic Station Juan Carlos I to study the behavior of NVIS communications at high latitudes. The NVIS nodes will collect data from nearby sensors and send it to the central node at the Antarctic Station. A network of NVIS nodes with routing protocols will extend the area of influence of the Antarctic Stations beyond the 250 km of a single hop.