Rev 29Aug03 GBT Receiver Upgrade Ideas with a Focus on Spectral Baselines The investigation of spectral baseline problems on the GBT has spurred us to think broadly about how to improve the wideband characteristics and stability of the entire signal path from the feed to the spectrometer output. We have a few changes in the works for improving the current system with regard to spectral baselines, but the full potential of the GBT for sensitive, broadband spectroscopy calls for a timely evolution to a fundamentally different receiver and signal transmission architecture. The current system covers a broad range of frequencies and is enormously flexible in signal routing and tunability, which will serve much of the science proposed for the GBT, but the necessary complexity and long, high frequency signal paths make it difficult to achieve the ultimate performance potential of the GBT. Below we outline a range of changes to the GBT receiver systems and the benefits expected from each. The over-all philosophy of the changes is to minimize the path length, the total gain, and the numbers of frequency conversions, connectors, and frequency-sensitive components between the feed input and the spectrometer samplers. We also propose a migration from complex and difficult to maintain special-purpose digital spectrometer hardware to a more flexible signal processing architecture based on general-purpose hardware that can more closely track the rapid improvement in digital technology and respond more quickly to changes in observer requirements. In the same way that the old technique of load switching has been replaced by total power spectroscopy as receivers have become more stable, we think that the current practice of equal observing time spent on signal and reference total power measurements can gradually be replaced by less frequent, low-noise system calibrations by moving the most frequency-selective components to digital hardware. Hence, the factor of two in the radiometer equation for signal/reference measurements may approach unity for modest bandwidth spectroscopy. May of the changes can be done incrementally resulting in improvements to GBT performance along the way. Some modifications are fairly straightforward while others involve more risk in terms of effort required to achieve the expected performance goals. Each project component can be assessed and managed largely on its own while adhering to the over-all system goals. Hence, there is minimal need for complex project management as long as the integrated plan is well specified at the outset and this plan is reviewed periodically to check for progress and coherence. Timely implementation of all improvements is very important to the quality of GBT science, however. Careful consideration needs to be given to the limited number of receiver slots on the GBT turret to avoid disrupting the flexibility of dynamic scheduling and the desire for infrequent receiver installations. Front-End Components Wideband spectral line tests on the GBT receivers have shown that there is considerable frequency structure in the gain and noise temperature of the feed, waveguide, and first amplifier stages of the front-ends. A bit of modeling of the measurable reflections in this part of the system shows that we should expect some variation in gain and noise temperature on the order of 10 to 30% of the observed effect as a combination of a few sinusoidal components that are characteristic of the delay distances in the feed. While these reflections are important, they do not fully explain the observed noise frequency structure either in amplitude or frequency scale-length. Careful investigations of the X-band and Ku-band feed/waveguide systems have shown that quite sharp but very weak resonances can exist and that they are quite sensitive to waveguide joint quality. A detailed look at our feed and waveguide construction techniques is called for. Can we improve or eliminate the mechanical joints between sections? Do the corrugated feeds have intrinsic resonances that make them less than optimum for spectroscopic work? Can we reduce the reflections and losses in radomes and dewar vacuum windows? Are there ways of suppressing resonances that cannot be completely eliminated? The optimum noise impedance match of the single-ended HFET amplifier is generally not the optimum power transfer impedance match. Hence, our current receivers have a built-in, substantial, power reflection at the first amplifier input. The higher frequency receivers use circulators at the LNA inputs to reduce these reflections, but there is considerable room for improvement in this area, particularly in the lower frequency receivers. Balanced and doubly-balanced amplifier designs show potential for improvement in this regard, both in terms of impedance match and in terms of thermal and amplifier noise radiated in the reverse direction. These designs need to be pursued and employed where they show significant advantages. The current feed/waveguide/amplifier structure seems too complex with a dewar window, thermal gap, OMT, waveguide-to-coax transition, coax with four connectors, cal coupler, and coax to microstrip transition. This combination is prone to mechanical stress and instability, RF loss, and reflections. A higher degree of integration is needed to include the cal coupler, input isolator, waveguide/microstrip transitions, and perhaps the OMT function to one assembly. The LNA design should be closely matched to the other components in the structure rather than designed to a generic input impedance. By reducing the length and number of components between the feed and first amplifier we should be able to eliminate a few Kelvins of system temperature caused by ohmic losses in this path. Fewer waveguide joints would mean fewer spurious resonances. Can we achieve adequate thermal isolation with a combination of foam, stainless steel, dielectrics, and reduced surface areas to eliminate the thermal gap and employ a lower-loss vacuum window? Compact IF System One signal path in the current GBT IF system goes through three to five frequency conversions, about 20 meters of coax carrying signals up to 6 GHz, dozens of connectors, considerably more amplifier gain than is necessary in many cases, analog fiber signal transmission, more than eight RF Selection switches, several variable attenuators, and several high-Q filters. Some of the constraints on this design, such as the necessity for analog signal transmission between the GBT and the lab, are no longer valid. We have paid a price, in terms of spectral baseline quality, for great flexibility in signal routing through an IF system that needs to serve all receivers from 300 MHz to 100 GHz. Digital modems for optical fibers can now handle most of the bandwidth requirements of the GBT below about 20 GHz sky frequency so the samplers should be moved to the GBT receiver room so that all signal transmission and routing is done digitally without loss of quality. This requires careful design and shielding of digital components to avoid RFI to the receivers, but this is well within our current capabilities. Wherever possible, the samplers should be integrated into each receiver package to avoid any IF cable runs. Careful consideration of the signal interfaces to existing back-ends will be required to plot a migration path from our current IF system to one that digitizes at the Receiver Room. Ultra-wideband spectrometers, such as the one being considered for short-wavelength spectroscopy, may require the use of an analog IF link for the foreseeable future, but there are dark fibers currently installed that would allow parallel operation of digital and analog signal paths. The IF module for each receiver should use an an absolute minimum of connectors consistent with good RF isolation and serviceability. Any cable lengths must be shorter than a few centimeters. Since each receiver will have its own IF electronics, it will generally need only two frequency conversions for adequate image rejection and passband selectivity before the sampler. The intermediate frequencies can be chosen without the constraint of conforming to a general-purpose IF system. One of the requirements of the current GBT IF system was that it be capable of simultaneously tuning to a number of spectral lines that are widely spaced in frequency. This requirement needs a careful look to identify the specific instances where such capability is required. Since we have this capability with the present system, development of new IF modules can proceed for awhile without this constraint, and it should not be too difficult to design this capability into certain receivers as needed. The main expense is in the number of second LO synthesizers. One of the goals of an new receiver system is to have a passband that has a minimum of gain ripple across the total power spectrum. Our present filters are designed for sharp cutoff at the edges of their passbands to maximize tuning range and useful spectrometer bandwidth while achieving good image and alias rejection. The anti-aliasing constraint can be relaxed by sampling faster than is required by the spectrometer bandwidth and doing the final stages of anti-aliasing with digital filters whose band shapes are perfectly stable and predictable. First IF filters can often be of lower Q when one is free to choose the IF center frequencies independently for each receiver. Zero-ripple filter types are preferred with careful design to minimize impedance mismatches at their input and output ports. A smooth passband reduces the need for frequent reference spectra in the observing strategy, which will save telescope time for modest bandwidth spectroscopy. Total power detectors for both accurate signal level monitoring at critical points in the amplifier chain and for continuum observing should be an integral part of the IF design. Where multi-bit samplers can be used, a wider dynamic range square law detector may be possible with digital processing of the A/D output. The variation in spectral power density in one receiver is generally less than 10 dB and often much less than this. The total system temperature changes less than about a factor of two, and continuum source antenna temperatures are usually much less than the system temperature. The RF gain across the receivers tuning range varies less than about 3 dB in most cases. Hence, the need for many stages of variable attenuation can be eliminated in a receiver-specific IF design. Our current general-purpose IF system has 64 dB of variable attenuation in two sets of 32 dB with one of the attenuators having 1/8th dB resolution. These require a total of 28 RF switches in the signal path which have been a significant source of reliability and repeatability problems. The spectral processor has an additional 64 dB of variable attenuation in 1 dB steps (12 RF switches) at its input. Experience has shown that even three-level samplers can tolerate 1/2 dB of input level variation with little loss in sensitivity, and multi-level samplers are considerably more tolerant. Sampler input power does depend directly on bandwidth, but this can be accommodated with fixed attenuators associated with each filter. Less variable attenuation means less total gain in the RF and IF stages, hence, better stability and less chance of amplifier non-linearity. With some attention to sampler sensitivity, the total system gain can be held to less than 80 dB and the amount of variable attenuation held to 8 or 16 dB in 1 dB steps. There are limitations of space, weight, and cooling capacity imposed by the GBT receiver room so the amount of IF and sampler electronics that can be moved into the receivers package will depend on our ability to design and build efficient modules. This will require some careful development work and the adoption of the most recent integration and fabrication techniques. Digital Signal Transmission Placement of the digital sampler in the GBT receiver room means that the analog optical fiber modems will need to be replaced with digital modems. The existing fiber can be used with the digital modems. A 10 gigabit per second modem is well within the 2003 state-of-the-art. This will carry more than 2 GHz of two-bit and about 500 MHz of eight-bit sampled IF bandwidth on each fiber. Since we have extra installed fibers, the digital and analog IF systems can coexist for any period of transition without much increase in signal routing complexity. Since no fidelity in the received signal is lost in the digital transmission path all digital signal processing hardware is best located in the shielded GBT Equipment Room in the Jansky lab. Some signal multiplexing hardware and possibly a few stages of digital filtering will be required in the GBT receiver room. Signal Processing In principle, the existing autocorrelation and FFT spectrometers can be used with the new receiver configuration and digital signal transmission by replacing the internal samplers with data streams from the digital fibers, but this needs a careful look. At the very least, sampler clock synchronization and data buffering to accommodate variations in propagation delay will be required. Each back-end requires different sample rates and resolutions, which were optimum cost/performance trade-offs at the time of their design. The true cost of accommodating legacy back-ends should be weighed against the cost and benefits of new signal processors. The long-term goal for GBT signal processing is to develop an array of independent, well-targeted, digital back-ends built around general-purpose processing hardware for various types of data acquisition. Special-purpose hardware, such as FPGA's and DSP's, that requires a lot of expertise to design and build will be confined to specific pre-processing tasks that require the greatest speed, such as digital filters and data multiplexers and demultiplexers. Much of the signal processing task will be migrated to software which is accessible to development by users of all types. Hopefully, this will generate interest in user-supplied algorithms for a wide range of data processing in the fashion that now exists in the pulsar community. One fundamental goal is to simplify individual signal processing back-ends so that the time from design concept to implementation is short enough to stay reasonably close to rapid developments is digital hardware speed and to be able to respond to observer requests for new capabilities in a timely fashion. Bandwidths as wide as the GBT autocorrelation spectrometer's 800 MHz are beyond the reach of current general-purpose hardware, but this will come. An example of today's state-of-the art is the Arecibo Signal Processor (ASP) developed by groups at Berkeley, Univ. of British Colombia, and Princeton. It has 32 input channels, each with 4 MHz bandwidth and 4-bit A/D samplers. These are attached to 2.4 terabytes of RAID disk space, 2 data servers, and 16 slave processors of 2.4 GHz Xenon cpu's in a Linux cluster. The slave processors can be programmed to do Fourier transforms, dedispersion, RFI excision or any other task that might be performed on the IF signal voltage samples. The processing power is probably capable of keeping up with a real-time IF bandwidth of about 20 MHz for FFT operations or the full 128 MHz in burst mode. This hardware architecture is amenable to expansion and upgrade to higher processor and bus speeds and larger disks as they become available. First Steps We cannot possibly attack all aspects of the suggested changes at once. A few areas need some careful R&D before specific designs can be implemented. One possible approach will be to select one or two receiver bands where the astronomical demand is heaviest or most challenging, where the a significant impact on GBT science can be made, and where the technical solutions appear to be within reach (wide-band spectroscopy at X-band or a very low-noise receiver system for L-band, for example). The current receiver system for that band should then be studied in considerable detail along the lines of the spectral baseline investigation to identify as many specific limitations in the current design as possible. A parallel R&D effort on critical components for the new architecture can proceed with a 12 to 16 month proof-of-concept goal for a new design based on the two efforts.