Broad-Bandwidth FPGA-Based Digital Polyphase Spectrometer

NASA’s Jet Propulsion Laboratory, Pasadena, California

With present concern for ecological sustainability ever increasing, it is desirable to model the composition of Earth’s upper atmosphere accurately with regards to certain helpful and harmful chemicals, such as greenhouse gases and ozone. The microwave limb sounder (MLS) is an instrument designed to map the global day-to-day concentrations of key atmospheric constituents continuously. One important component in MLS is the spectrometer, which processes the raw data provided by the receivers into frequency-domain information that cannot only be transmitted more efficiently, but also processed directly once received. The present-generation spectrometer is fully analog. The goal is to include a fully digital spectrometer in the next-generation sensor. In a digital spectrometer, incoming analog data must be converted into a digital format, processed through a Fourier transform, and finally accumulated to reduce the impact of input noise. While the final design will be placed on an application specific integrated circuit (ASIC), the building of these chips is prohibitively expensive. To that end, this design was constructed on a field-programmable gate array (FPGA).

The measured channel frequency response from the 8192-channel **Spectrometer**.

A family of state-of-the-art digital Fourier transform spectrometers has been developed, with a combination of high bandwidth and fine resolution. Analog signals consisting of radiation emitted by constituents in planetary atmospheres or galactic sources are down-converted and subsequently digitized by a pair of interleaved analog-todigital converters (ADCs). This 6-Gsps (gigasample per second) digital representation of the analog signal is then processed through an FPGA-based streaming fast Fourier transform (FFT). Digital spectrometers have many advantages over previously used analog spectrometers, especially in terms of accuracy and resolution, both of which are particularly important for the type of scientific questions to be addressed with nextgeneration radiometers.

The high-level building blocks (filter and FFT components) were optimized for the Xilinx Virtex 5 FPGA, and for interfacing with one another. The design, from building blocks to complete implementation, was floor-planned in order to make efficient use of the FPGA resources. As more aggressive spectrometer designs were created, designing the hardware to run at a sufficiently high clock rate became progressively more difficult. These issues were mitigated by duplicating hardware and adding (or removing) latency as necessary. The floor-planning of the design was changed dramatically from the original.

The final spectrometer design is an 8192-channel implementation. Designed with additional output capacity, the spectrometer has superior frequency resolution, dynamic range, and accumulation length when compared to previous versions. An alternate, dual-polarization, 1.5-GHz, 4096-channel spectrometer is available as well. Both designs are capable of accumulating for hours, several orders of magnitude over what is required.

In addition, a further improved spectrometer with double the frequency resolution, a polyphase-FIR filter front end, and substantially reduced noise has been successfully simulated and is presently in the final stages of development. When finished, it will offer a spectrometer developed on Virtex-5 hardware with bandwidth and spectral resolution an order of magnitude greater than the analog spectrometers presently in use.

Plans to make an 8-GHz spectrometer taking advantage of the same technology used for this device are already being made. Finally, efforts are presently being made to interface this design to a compact Nallatech board, which consumes less power and can be more readily used in remote locations and demanding environments.

This work was done by Robert F. Jarnot of Caltech and Ryan M. Monroe of Georgia Tech for NASA’s Jet Propulsion Laboratory. NPO-48352