Satellite Monitoring Solutions | Teledyne SP Devices

Home Applications Satellite Monitoring

The goal of satellite monitoring is to assure uplink/downlink quality, interference detection, and compliance with spectrum regulations. The systems involve three main components:

1. Space: Satellites equipped with transponders and antennas operating in specific frequency bands
2. Ground: Monitoring stations with large parabolic antennas, digitizers, and RF front-end systems to capture uplink and downlink signals.
3. User: End-user terminals or specialized monitoring equipment that process and analyze the received data.

Satellite communication uses designated RF bands, each with unique characteristics. The image below shows selected bands.

Figure 1. Examples of satellite bands and their usage.

Each frequency band is divided into sub-bands that operators use to allocate spectrum for specific services. In bidirectional systems, these sub-bands are split into uplink (earth to satellite) and downlink (satellite to earth) to prevent interference.

Typically, the lower part of the band is reserved for downlinks because it experiences less atmospheric attenuation, making reception easier. The higher part is used for uplinks, which support higher data rates and avoid interference with downlink signals.

Sub-band names, center frequencies, and instantaneous bandwidths vary across different systems such as the Global Navigation Satellite Systems (GNSS). For example, the European Galileo system uses “E” designations instead of “L” within the L-band, reflecting its internal frequency plan. The illustration below shows Galileo downlink sub-bands as an example.

Figure 2. Galileo sub-bands in the L-band.

Ground stations capture radio frequency (RF) signals transmitted by satellites using high-speed data acquisition boards, or digitizers, that convert analog RF signals into digital data for processing. These digitizers often support direct sampling of L- and S-band signals without mixers, reducing system complexity and cost. Additionally, field-programmable gate array (FPGA) real-time processing such as digital down-conversion and demodulation may be utilized in order to prepare the data for storage or transmission for further analysis.

Data Acquisition

The input bandwidth of the digitizer needs to be sufficiently high in order not to attenuate signal frequency components in the band of interest. Additionally, external signal amplification may be required in combination with mandatory anti-alias filters that remove out-of-band components and prevent aliasing – a type of distortion that occurs when high-frequency components fold into the band of interest during analog-to-digital conversion. Amplifiers and filters with SMA connectors are available as commercial-off-the-shelf components for specific bands such as the L-band.

Frequency planning and choice of appropriate sampling rate are important when selecting a digitizer. It is crucial to ensure that the signal of interest falls within a single specific Nyquist band, and that out-of-band signal components are suppressed by filtering. Sampling signals in the first Nyquist band is referred to as baseband sampling, while sampling signals in higher Nyquist zones is referred to as bandpass sampling or undersampling. The minimum sampling rate should be at least twice the signal bandwidth, provided proper bandpass filtering is applied, and this translates to 2, 4, and 8 Giga-samples-per-second (GSPS) for L-band, S-band, and C-band, respectively.

The ADQ35-WB digitizer from Teledyne SP Devices offers 12-bit resolution, 9 GHz usable input bandwidth, and 3, 4, or 5 Giga-sample-per-second (GSPS) sampling rate in dual-channel mode or 6, 8, or 10 GSPS in single-channel mode. This allows for direct sampling of the full L- and S-bands. Digital equalization can be utilized for C-band sampling to compensate for analog bandwidth roll-off.

Figure 3. Teledyne SP Devices’ ADQ35-WB digitizer enables direct sampling of satellite frequency bands and offers an onboard open FPGA for custom digital signal pre-processing.

The sampling rate options offer great flexibility in optimizing frequency planning. For example, the L-band is preferably sampled at 5 GSPS and will fall within the first Nyquist band of the digitizer. S-band undersampling is best achieved using 4 GSPS sampling given that this provides a sufficient margin for filter passbands (since the S-band covers the entire second Nyquist zone). Note, however, that the 5 GSPS sampling option would not be a good choice as the signal of interest – between 2 and 4 GHz - would then span both the first and second Nyquist zones (split at 2.5 GHz) which would inevitably introduce aliasing. For the same reasons, C-band undersampling is best achieved using 8 GSPS.

Figure 4. ADQ35-WB typical frequency response (left) and two-tone test (right).

The benefit of selecting a sufficiently high sampling rate is that it limits the output data rate, which helps simplify post-processing. Alternatively, a higher sampling rate can intentionally be selected, given that the signal of interest does not cross Nyquist zone boundaries, and then combined with digital post-processing such as decimation in order to achieve signal-to-noise (SNR) improvements while still reducing data rates for post-processing. This is covered more in-depth in the coming sections.

FPGA Pre-Processing and Data Transfer

FPGA pre-processing is essential in satellite monitoring, where the data volume from high-speed digitizers can be overwhelming. The ADQ35-WB performs 10 billion measurements per second, and with each sample represented by 2 bytes, this translates to an astonishing 20 GB/s data rate. At that pace, a 1 TB (Terabyte) SSD fills up in just 50 seconds!

For high-throughput applications like satellite monitoring, digitizers with a PCIe interface are preferred. The ADQ35-WB supports transfers to the host PC at speeds up to 14 GB/s, but the raw 20 GB/s output still needs reduction to avoid saturating the PCIe link. This pre-processing step not only prevents bandwidth bottlenecks but also simplifies downstream post-processing.

FPGA Data Reduction Examples

Countless real-time data reduction methods can be implemented in the open onboard FPGA, but two types stand out as suitable candidates for satellite monitoring:

Bit compression: ADQ35-WB features built-in bit compression that allows the user to transfer each measurement value as, for example, 10 bits. This corresponds to 1.25 bytes per sample, and hence a total data rate of 12.5 GB/s from each digitizer. This is below the 14 GB/s PCIe link capacity and can therefore be streamed continuously. The benefit of this method is that the full L- or S-band acquisition can be transferred without any data loss for either additional post-processing or recording to disk.
Digital Down Conversion: The FW2DDC firmware option equips the digitizer with two real-time digital down converters (DDCs) inside the onboard FPGA. The DDCs enable frequency translation and mixes down the RF signal to baseband or a lower intermediate frequency using a numerically controlled oscillator (NCO). It also contains filters to isolate the frequency band of interest as well as decimation which is a technique that can be utilized to reduce the data rate in order to simplify potential subsequent post-processing. Additionally, the combination of filtering and decimation also helps improve the signal-to-noise ratio (SNR). FW2DDC supports a wide variety of radio architectures with either real-valued or complex inputs, such as in-phase and quadrature (IQ) signals.

Example: Direct sampling of Galileo signals

Figure 5. The Galileo sub-bands span 427 MHz.

Using dual-channel mode allows for acquisition of two polarizations, and 5 GSPS is an excellent choice of sampling rate when combined with digital down conversion. Note that the entire 427 MHz signal fits within the first Nyquist zone at up to 2.5 GHz.

By tuning the NCO (mixer) frequency in FW2DDC to -1377.5 MHz, the band is down converted to 0 and represented as I and Q that span the range -213.5 MHz to 213.5 MHz.

The data rate can be reduced and the SNR improved by using a combination of decimation and filtering. Decimation by 8 yields a sampling rate of 625 MSPS and the subsequent FIR filter offers an 80% passband relative to Nyquist (0.8 x 312.5 MHz = 250 MHz). This improves SNR by approximately 9 dB due to reducing the effective noise bandwidth, and the resulting data rate is 625 MSPS x 2 bytes x 2 channels = 2.5 GB/s. This data rate can be streamed to a GPU for further down conversion and demodulation.

Figure 6. Decimation combined with filtering helps reduce data rate and improve SNR.

Peer-to-peer Data Transfer

Conventional data transfer routes digitizer output data through the host PC's CPU and system memory before reaching endpoints such as graphics processing units (GPUs). This indirect path introduces latency, consumes valuable CPU resources, and restricts throughput due to memory bandwidth limitations.

Figure 7. Conventional streaming involves extensive read/write operations to/from multiple memory segments.

Peer-to-peer (P2P) streaming, by contrast, uses PCIe Direct Memory Access (DMA) to transfer data directly from digitizers to GPUs. By bypassing the CPU and RAM entirely, P2P eliminates these bottlenecks and enables true high-speed performance. Multiple digitizers can stream simultaneously to a single endpoint, achieving aggregate data rates of up to 56 GB/s on PCIe Gen5 x16, making it ideal for real-time processing and large-scale data capture.

Figure 8. With peer-to-peer streaming, one or more digitizers can transfer data directly to GPU memory.

Post-Processing and Disk Recording

Onboard FPGAs are crucial for reducing the data rate in order to match the PCIe link capacity. The user can utilize application-specific add-on firmware or add custom real-time signal processing via the open FPGA. However, resources are limited, and additional subsequent post-processing may therefore be required.

GPU Processing

GPUs provide cost-effective post-processing for a wide range of signal processing tasks. Compared to FPGA development, creating and maintaining software for GPUs is typically faster and easier. GPU-based post-processing further reduces the data rate before storage, easing both transfer bandwidth requirements and overall storage capacity.

For example, a GPU-based channelizer can extract Galileo sub-bands from the combined 427 MHz signal acquired via direct sampling, as shown in the example above. After decimation, the data rate is reduced to about 2.5 GB/s—well within the capabilities of modern GPUs, where PCIe Gen5 interfaces can handle up to roughly 56 GB/s.

The ADQ35-WB supports consumer-grade GPUs as well as high-performance professional models, such as the NVIDIA RTX 4500, delivering extreme processing power for demanding applications.

Figure 9. Data acquisition system with digitizers, NVIDIA RTX4500 GPU, and SSD modules.

High-Speed Disk Recording

PCIe systems requiring high-speed storage often use RAID arrays built with PCIe-to-NVMe adapter boards hosting multiple NVMe SSDs. These boards employ bifurcation to split a single x16 PCIe slot into multiple independent sets - for example, x4/x4/x4/x4 for four NVMe drives - enabling parallel write operations across separate PCIe lanes. Alternatively, carrier boards with an onboard PCIe switch can be utilized. Regardless of the method, the aggregate throughput scales linearly with the number of drives.

Sustained write speeds depend on the NVMe modules used, and writing raw data blocks directly to disk sectors (bypassing the OS file system) maximizes efficiency. This is supported by Teledyne SP Devices' proprietary NVMe streaming library, libads. Both consumer and enterprise-grade SSDs can be utilized and PCIe Gen4 typically supports up to 7 GB/s per disk or 14 GB/s with PCIe Gen5. However, consumer drives suffer performance drops during long-duration recordings due to SLC (Single-Level Cell) cache limitations. They still remain cost-effective for short bursts - up to a few hundred Gigabytes - before cache exhaustion occurs and significantly reduces achievable write speeds.

By contrast, enterprise-class drives maintain high sustained throughput indefinitely. Configurations using these drives can reach aggregate write rates of 56 GB/s and a total storage capacity of 1 PB (Petabyte) per slot.

Figure 10. PCIe-to-NVMe adapter board with four Kioxa CD8 NVMe enterprise-class SSD modules supporting 15 Terabytes per drive.

Conclusion

PCIe-based multi-channel data acquisition systems deliver a cost-efficient, high-performance solution for satellite monitoring. Onboard FPGAs allow users to execute real-time, computationally intensive signal processing directly on the raw data stream, and high-speed PCIe links transfer output to multiple endpoints.

Cost-effective post-processing on GPUs enables parallel handling of data from multiple digitizers, simplifying workflows and reducing development time. For storage, commercial off-the-shelf enterprise-grade SSDs support aggregate disk streaming up to 56 GB/s with total capacities of 1 PB per slot, while more economical options remain viable for short-duration, full-speed recording.

Figure 11. PCIe-based satellite monitoring system using two ADQ35-WB digitizers with 25 GB/s write speed and a total storage of 75 TB.

Learn more on our website using the links below.
DPS7-PCIe multi-channel data acquisition system:
https://www.spdevices.com/what-we-do/products/hardware/multi-channel-data-acquisition/dps7-pcie

Disk streaming application note:
https://www.spdevices.com/en-us/Products_/Documents/ADQ35/25-3199 Disk streaming solutions with libads.pd

Model	Resolution	Channels	Sampling Rate	Coupling	Input Bandwidth	ENOB max	Interface
ADQ7DC	14-bit	1 2	10 GSPS 5 GSPS	DC	3 GHz	9.1 bits	PCIe, PXIe, USB 3.0, MTCA.4, 10GbE
ADQ7WB	12-bit	2	5 GSPS	AC	6.5 GHz	8.7 bits	PCIe, PXIe
SDR14TX	14-bit	2	2 GHz	DC	Up to 2 GHz	N/A	PCIe, PXIe
ADQ35-WB	12-bit	1 2	10 GSPS 5 GSPS	AC	9 GHz	8.8 bits	PCIe, USB 3.2