End-to-End Precoding Validation Over a Live GEO Satellite Forward Link

In this paper we demonstrate end-to-end precoded multi-user multiple-input single-output communications over a live GEO satellite link. Precoded communications enable full frequency reuse schemes in satellite communications to achieve broader service availability and higher spectrum efficiency than with the conventional four-color and two-color reuse approaches. In this scope, we develop a novel over-the-air test-bed for end-to-end precoding validations. We were able to demonstrate precoded communications over an actual geostationary orbit multi-beam satellite using the DVB-S2X standard-compliant state-of-the-art gateway and user terminals. The developed system is capable of fully end-to-end real-time communications over a satellite link and demonstrates novel channel measurements and channel compensation techniques for differential frequency and phase tracking. It is shown, that by successfully canceling inter-user interference in the actual satellite link, the precoding improves the received SINR and increases the total system goodput in aggressive full frequency reuse scenarios.


I. INTRODUCTION
The new era of broadband internet and on-demand services has inspired new approaches towards the design of satellite communications (SATCOM) systems.The market importance of broadband services and the limited frequency resources drive the SATCOM industry and academia towards the development of more efficient communications.Multi-beam satellites, on the one hand, are more powerefficient and, on the other hand, have higher capacity in the satellite channel through the spatial multiplexing [1].While conventional multi-beam systems employ the four-color (4CR) or two-color (2CR) reuse schemes to mitigate interference between the beams, precoding enabled full frequency reuse (FFR) schemes are more efficient in terms of spectrum utilization.Therefore, the application of multi-user The associate editor coordinating the review of this manuscript and approving it for publication was Adao Silva .
multiple-input single-output (MU-MISO) transmission in forward link SATCOM is highly challenging due to the practical constraints, but at the same time, extremely rewarding [2], [3] from both academic literature and industrial project point of views.
Fig. 1 shows the 4CR scheme for multi-beam satellite systems.In the usual case of a 4CR scheme, interference is alleviated by allocating two different carrier frequencies or two polarizations to adjacent beams.In this scenario, the satellite is capable to efficiently mitigate interference between user terminals (UT) in different beams.
Fig. 2 shows the 2CR scheme.In this scenario the number of the spectrum resources required is reduced twofold, but the interference will occur in the adjacent beams with the same color (frequency or polarization).
Fig. 3 shows the FFR scheme.The scenario requires only a single color to operate, but the amount of interference in UTs from adjacent beams becomes substantial.Advanced signal   processing techniques are required to manage interference.Recent works study the practical application of precoding in SATCOM [3]- [7].Precoding can be implemented at the gateway in the digital domain.For precoding to be efficient, each UT has to accurately estimate channel state information (CSI) and report it to the gateway.
To facilitate industrial adaptation, extensive in-lab and field tests are required to increase the technology readiness level.In [8]- [10] the authors presented an in-lab real-time satellite precoded transmission test-bed, where a satellite has 6 transmitting antennas and simultaneously serves 6 user terminals with up to 32 Amplitude and phase-shift keying (APSK) modulated signals.Nevertheless, the demonstrated test-bed was developed for the in-lab environment and tested over the simulated multi-beam satellite channel, which considered a near-instantaneous channel estimation feedback at the gateway side and did not include the additional impairments of differential phase noise at the satellite payload.The authors in [11], [12] showcase over-the-air validations using actual satellite links by using closed-loop Zero-Forcing.The authors use a custom framing for synchronization and estimate signal-to-interference-plus-noise ratio (SINR) gains at the receivers.The authors in [13] demonstrate their field trials of multi-satellite multi-user precoding.In their work, they validate precoded communications over two satellites and conduct the trials by using a modified Digital Video Broadcasting -Satellite (DVB-S2) signal modulator with custom pilots and reference tones injection.However, the fact that two separate satellites are used did not allow the study of differential effects between two RF chains of the same satellite payload.
In this work, we demonstrate a complete end-to-end test-bed for precoding validation over a live satellite link.The test-bed consists of a DVB-S2X standard-compliant [14] gateway and a setup of two UT.In the gateway we implement two types of closed-loop minimum mean-squared error (MMSE) precoding to mitigate the inter-user interference.The gateway as well uses a closed-loop approach to compensate differential frequency and phase between the beams induced by the different local oscillators of the satellite transponders.The differential frequency and phase values are derived from the actual CSI data.The CSI is estimated at the UTs side using the standard pilots present in the superframe structure.Thus, the gateway is not required to be located in the coverage area of the beams to compensate for the impairments.Also, by using the standard pilots to estimate all the considered channel impairments the system does not use excessive signaling or bandwidth to operate.Using the developed test-bed we validate the end-to-end throughput performance of precoded communications in the realistic environment.Therefore, the main contribution reported in this work are • The DVB-S2X with Annex E superframe format specification 2 standard compliant gateway and user terminals are presented.
• Real-time channel measurements and compensation of the channel impairments are described.
• End-to-end precoding validation over a live satellite forward link are demonstrated.
• Performance in terms of SINR and goodput at the UTs is measured using the precoded FFR communications.
The remainder of this paper is organized as follows.In Section II, we describe the over-the-air (OTA) setup.OTA results are shown in Section III, followed by concluding remarks in Section IV.
Notation: Upper-case bold-faced letters are used to denote matrices.The superscripts (•) † and (•) −1 represents matrix transpose and inverse operations, respectively.| • | is an absolute magnitude of a complex value.

II. TEST-BED SETUP DESCRIPTION
Fig. 4 shows the overview of the precoding validation setup, established for the precoding validation over a satellite link.In this setup, the gateway is serving two single-antenna UTs simultaneously over the satellite forward link.Each UT receives its corresponding spot beam and the interfering spot beam at the same time.We use the standard DVB-S2X [14] superframe format 2 for the satellite forward link communications.The UTs estimate the CSI based on the DVB-S2X frame embedded pilots and report the estimated values to the gateway through a dedicated feedback channel over a landline Ethernet link.The gateway uses this CSI data to compute a precoding matrix, which is applied to the transmitted signals to cancel the inter-user interference.
The use cases of modern SATCOM systems in 5G networks for providing off-load backhauling for unicast user traffic [15], [16] allow scenarios where an Ethernet link over the terrestrial and landline network is commonly available.In the hybrid satellite-terrestrial networks for maritime [17] the CSI feedback can be established through a satellite or a terrestrial Ethernet link, according to the availability.The standalone broadband ground stations can initiate a satellite up-link connection [18] and feedback the CSI data.
In in this work, we consider a 2×2 MU-MISO system.The received signal at the i-th UT is given by y i = h † i x+σ i , where h † i is a 1 × 2 vector representing the complex channel coefficients between the i-th UT and the 2 satellite transponders, x is defined as the 2 × 1 vector of the precoded transmitted symbols of a DVB-S2X superframe at a certain symbol period and σ i is the thermal noise measured at the i-th terminal's receive antenna.Looking at the general formulation of the received signal, which includes the whole set of terminals, the signal model is where y ∈ C 2×1 , σ ∈ C 2×1 , x ∈ C 2×1 , and s ∈ C 2×1 and H ∈ C 2×2 .

A. SATELLITE SPOT BEAMS
For the satellite link we use SES-14 satellite.SES-14 is operational since 2018.The hybrid satellite provides C-and Ku-band wide beam coverage, as well as Ku-and Ka-band High Throughput Satellite (HTS) coverage across the Americas and the North Atlantic region.Also, SES-14 has two spot beams in the Ku-band covering a part of Western Europe and the United Kingdom.We show the approximated spot beams in Fig. 5.In these validations, we use two channels (one per spot beam) in Ku-band with 12.4 MHz bandwidth.We consider the following design solution of the UTs and the gateway to conduct precoding validations over the proposed satellite and the spot beams.First, the signals from the gateway for the two spot beams are relayed by the satellite through two independent transponders.The transponders have undisciplined local oscillators, which results in differential frequency and phase offsets between the spot beams, received by the UTs.This impairment can substantially reduce the precoding performance.Therefore, we implement a frequency and phase tracking and compensating loop in the gateway to minimize these offsets.Second, the spot beams are separated in the vertical (V) and horizontal (H) polarization.The UTs are placed in two locations where they receive the signals from both spot beams and polarizations.Then, the OTA interference channel is generated by polarization mixing in the digital domain of the signal processing chains at the UTs.
We discuss in detail the design and setup of the gateway and UTs in the following sections.

B. GATEWAY DESIGN AND SETUP
We use the commercially available USRP (Universal Software Radio Peripheral) 2974 developed by National Instruments (NI) for a RF front-end and deploy our custom software.The USRP consist of 2 transmitting and 2 receiving channels and an integrated FPGA module Kintex-7 from Xilinx.We implemented a real-time DVB-S2X transmitter in the FPGA of the USRP as shown in Fig. 6.
The transmitter consists of in-house build FPGA blocks.The FPGA blocks generate 2 streams of symbols of the  DVB-S2X superframe as shown in Fig. 7.The superframe structure contains regularly inserted and aligned between spot beams unprecoded SF-pilots and precoded P2 pilots.The SF-pilots, with the length of 36 symbols, are used for channel estimation at the UT side.The pilots transmitted on each beam are orthogonal Walsh-Hadamard sequences.The P2 pilots are 180 modulated symbols located at the beginning of each bundle frame.These pilots are used to estimate SINR.Since the P2 pilots are precoded with the same precoding matrix as the data symbols in the frame, the resulting SINR is relevant to all the data symbols in a corresponding frame.Each DVB-S2X stream carries terminal-specific data.The data is encoded with the Low Density Parity Check (LDPC) forward error correction (FEC) codes procedure [19].The streams are jointly precoded by the PRECODER block, which applies the precoding matrix W on the symbols.The block is configured to precode only certain segments of the DVB-S2X superframe.The block does not precode Start of superframe (SOSF) and SF-pilots.The SOSF is a known sequence, which can be reliably detected at a user terminal even in a high interference environment.The P pilots are not precoded because they are used by the UTs to estimate CSI ( Ĥ) and calculate differential frequency and phase offset between the two spot beams.
There are two types of MMSE precoding implemented at the gateway.The first techniques is the regularized MMSE [20], which can be defined as where σ -noise variance measured at the UT side, and Iidentity matrix.
The second MMSE per-antenna power constrained (MMSE PAC) technique was presented in [21] is defined as where is a real diagonal matrix consisting of Lagrangian dual variables.The goal is to find the optimal regularization factor, which should satisfy where φ is the available power level at each transmitting antenna.The authors in [21] have proposed an iterative method to find and proved its convergence.Due to the requirements of the gateway feeder link for constant output signal we normalize the output magnitude per each antenna.The magnitude per-row upper bound of the precoded signal is considered |w1,1|+|w1,2| ≤ 1 for the normalized QPSK and PSK case.The precoding matrix for the 2 × 2 system can be then normalized as where The modulated symbols, generated by the gateway before the precoding is applied are normalized.Therefore, the normalization is only required for the precoding matrix itself.The authors would like to acknowledge, that the current gateway feeder link limitations will affect the performance of the precoders.Even though the current satellite infrastructure was not designed to support precoding, we show that it is still capable to benefit from precoded FFR communications.Finally, the gateway USRP is connected to the feeder link as shown in Fig. 8.We use an additional host PC to control and monitor different parameters of the USRP during the operation.

C. USER TERMINAL BLOCK DESIGN AND SETUP
The UT design is shown in Fig. 9.We use the same model USRP 2974 as for the gateway design.The UT is capable to receive and decode the DVB-S2X superframe in a real-time regime.The UT decodes the data symbols using the LDPC decoder, calculates the frame error rate and data rate performance.
We set up two locations for the UTs as shown in Fig. 10.In these locations each UT is connected to a dual-polarization Low Noise Block converter (LNB) as shown Fig. 11.While commercially available receivers provide power to an LNB, we use an external DC supply through the DC bias-tee.The host PC is used to control and monitor the USRP.
The dual-polarization LNB outputs two signals, each corresponding to the vertical and horizontal polarizations of the spot beams.We implement a signal mixing of the signals from the two spot beams in the digital sampled domain of the USRP.The block diagram of this setup is shown in Fig. 12.The signal mixing emulates the FFR scenario, where both spot beams are continuously transmitting using the same    carrier frequency and polarization of the forward link channel causing strong inter-user interference.
The pre-aligned LNB outputs the V and H signals on two different RF channels, which are connected to the two RF inputs of the UT.With this setup, we can control the ratio of the two signals with the parameters α and β.The received power at the UT after combining can be then expressed as: where P N is the combined thermal noise of the LNB and USRP RF front-end.The equivalent of the introduced signal mixing is receiving both signals combined already at the same RF output of the LNB.This could be achieved if the LNB was purposely re-aligned to achieve cross-polarization mixing of the signals.If we rotate the LNB by the angle with respect to the satellite polarization the received power at the UT would be: We can see that the both signal mixing techniques yield identical received signal for = 45 • and α = β = 0.5 (mixing in equal parts) The UTs use the SF-pilots of the superframe format 2 to estimate CSI.Additionally, UTs can calculate and report the differential frequency and phase between the two spot beams induced by the satellite payload.The authors in [9] demonstrated degradation in the performance of precoding over multi-beam satellites with independent local oscillators.We implement a differential phase and frequency tracking loop to minimize this impairment.
The P pilot k-th sequence P pilot k is generated as in [14].The pilot fields are determined by a WH sequence of size 32 symbols plus padding.The selection of the parameter k is a static choice for the transmit signal, thus we transmit a known sequence to the UTs on each spot beam.The pilot sequences are multiplied by (1 + ι) 2 to generate modulated symbols.We can distinguish each orthogonal WH sequence in the jointly received signal and estimate CSI for the k-th received beam at the receiver side as where y[t] is the received signal and N WH is the length of the WH sequence.
To calculate the differential phase and frequency from the CSI data we apply the approach presented in [22]: where T is the symbol period.The differential phase compensation requires a minimal delay between the phase estimation at the UT side and the correction applied at the GW side.We achieve the best performance when the phase sensing is performed directly at the GW rather than at the UT side.Since the UTs do not share the same clock reference with the gateway for the local oscillators, we consider the phase of one beam as a reference and calculate the phase difference to the other.Therefore, we track only the differential phase between the spot beams and compensate it at the gateway side.
To evaluate the performance of the precoding techniques, we consider the SINR measured from the P2 pilots at the UT side.The DVB-S2 implementation guidelines [19] describe the Signal-to-Noise Ratio Estimation (SNORE) algorithm.
The SNORE algorithm works in data-aided (DA) operating mode.At each UT, SINR is calculated as follow: 41560 VOLUME 11, 2023 Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
where P S -power of the intended signal and P N -power of noise plus interference.We calculate P s as where N P2 is the total number of symbols in the P2 pilots sequence P P2 .We calculate P N from the total signal power where The value of P N = σ 2 is also reported to the GW and used to calculate precoding in (2).

III. OVER-THE-AIR VALIDATIONS
For these over-the-air validations, we consider the emulated FFR scenario.In the scenario, both UTs are receiving signals from both spot beams simultaneously.The inter-user interference is mitigated through CSI estimation and precoding of the transmitted signal.First, we evaluate differential phase compensation and measure CSI data over a long period of time, and assess the CSI stability in both phase and amplitude after the compensation is applied.Finally, we measure the actual SINR and coded goodput reported by the UTs under the strong interference scenario with and without precoding applied on the transmitted signals.

A. DIFFERENTIAL FREQUENCY AND PHASE COMPENSATION
We show in Fig. 13 and 14 the time series of the differential frequency and phase measured between the two spot beams before and after the compensation loop was enabled at the gateway side.We can see that without the compensation the receiver experience a considerable differential frequency offset between the spot beams.Therefore, the phase difference is substantially distorted.It is evident that the received spot beams are not received coherently without the compensation loop.After we enable the compensation loop the differential frequency measured at the receivers is reduced below 1 Hz.As the result, the variation of the differential phase is substantially reduced.The measured phase difference between any two sequential points in time is less than 20 degrees.

B. FIELD MEASURED CSI
In this section, we collect and analyze the CSI data, measured at the UT0 and UT1.
Fig. 15 shows the measured relative magnitudes of the Horizontal polarization (H-pol) and Vertical polarization (Vpol) spot beams at the UT0 during 2 hours time series.We can see that the relative magnitude of both spot beams measured by the UT is varying within 1 dB precision.This variation occurs due to the precision of the estimated CSI at the UT.We also plot the mean of the magnitude of both spot beams and see that the variation is much smaller even for the long measurements.Fig. 16 shows the measured relative magnitudes of the H-pol and V-pol spot beams at the UT1 during the same 2 hours time frame.We see a similar variation of the measured magnitude of the spot beams.The average of the over-the-air collected data indicates the slowly varying satellite channel, which is expected for a GEO stationary satellite.The measured levels of the spot beams at the UTs stayed the same during the precoding validations conducted in this work.

C. PRECODING GAINS
The mixing of the polarizations introduced at the UTs enables validations of an emulated FFR 2 by 2 MU-MISO system.First, we measure SINR at the both terminals without precoding and show it in Fig. 17 and 18.The UT0 is receiving its data from the V-pol beam, thus the H-pol beam is introducing the interference.The H-pol beam is stronger than the V-pol beam in the location of the UT0, resulting in low SINR for this terminal as visible in Fig. 17.The UT1 is receiving its data from the H-pol beam.In its location, the H-pol beam is around 5 dB stronger than the interfering V-pol beam, which results in much better SINR at the terminal, which we can see in Fig. 18.
In the next step, we enable the MMSE precoding for the same strong interference scenario and measure the new SINR values at the terminals.In Fig. 19 we see that the SINR at the UT0 has improved from around −2 dB to 4.2 dB on average.In Fig. 20 we see the SINR of the UT1 has decreased from around 5 dB to 3.6 dB on average.In Table 1 we summarize the SINR values at the UTs with MMSE and MMSE PAC techniques and also without precoding.For the MMSE PAC precoding in this scenario, the SINR at the UT0 was measured at 4.3 dB, while the SINR at the UT1 at 3.7 dB on average.During the overthe-air tests, we registered substantial variations of the SINR measurements during a single measurement as can be seen in Fig. 19 and 20.Thus, we need to further improve the SINR measurement to be able to reliably conclude the difference between the performance of MMSE and MMSE PAC precoding techniques.Nevertheless, the changes of the SINR values while switching from the unprecoded case to the MMSE case are very evident.
To fully validate the enhanced system performance while using the precoded communications, we measure the performance gains in terms of achieved goodput at the two UTs and the total system gain.In Table 2 we show the achieved coderates (CR) and goodputs (GP) for the precoded and unprecoded scenarios.In the unprecoded case, the UT0 has negative SINR which does not support the lowest coderate, implemented in our system.With precoding enabled, the UT0 has a much higher SINR to support the given coderate.The SINR of the UT1 was reduced, forcing it to use a lower coderate.Thus, the overall system throughput is improved by 48 %, while utilizing the same total system bandwidth for the forward link communications.We can see that the improved SINRs were translated into increased system throughput when using the precoding techniques.

IV. CONCLUSION
In this work, we demonstrated novel end-to-end precoded communications over the actual geostationary orbit satellite link.To the best of the authors' knowledge, this is the first demonstration of the standard DVB-S2X superframe compliant terminals, which can estimate in real-time CSI to facilitate precoding at the gateway.Using the superframe structure the gateway is capable to compensate the differential frequency and phase introduced by the conventional satellite transponders design.In the conducted field test, we demonstrated end-to-end SINR and coded goodput gains of precoded communications over the actual satellite forward link.The gains show that terminal-specific data can be transmitted to the independent user terminals through the same physical channel by utilizing closed-loop precoding over a multi-beam satellite.It is shown, that precoding techniques enable FFR communications in SATCOM over the conventional color reuse schemes.

FIGURE 4 .
FIGURE 4. Overview of the experimental setup.

FIGURE 8 .
FIGURE 8. Gateway connection to the feeder link.

FIGURE 11 .
FIGURE 11.UT connection to the LNB.

FIGURE 13 .
FIGURE 13.Time series (25 seconds) of the differential frequency and phase measured between the spot beams with disabled compensation at the gateway.

FIGURE 14 .
FIGURE 14.Time series (25 seconds) of the differential frequency and phase measured between the spot beams with enabled compensation at the gateway.

FIGURE 15 .
FIGURE 15.Measured relative magnitude of the V-pol and H-pol spot beams at the UT0.

FIGURE 16 .
FIGURE 16.Measured relative magnitude of the V-pol and H-pol spot beams at the UT1.

TABLE 1 .
SINR gains at the terminals.