The specifications for the Polarimeter have been shown in the following table. These were taken after considering several optimisations. In-depth studies of Pulsar polarizations demands full polarization information ( meaning, the linearly polarized, circularly polarized and unpolarized intensity components of the pulsar radiation ) with a time resolution limited only by the recording capabilities. Thus, the polarimeter is designed as a separate module, capable of producing polarization information in the terms of Stokes parameters at full speed, limited only by the raw bandwidth of the receiver. The incoming data is accepted in the form of 256 spectral channels, in time-multiplexed form with one complete spectrum every 16µs. The input section includes a frame-selection logic, which can be programmed to drop a given number of spectrum frames between every pair of accepted spectra. The Stokes parameters are routed to the DSP nodes through a flexible distribution logic, which splits the spectrum into groups of channels and routes the groups to one or more nodes, in a user defined sequence. The number of channels supplied per node and the number of nodes selected to handle all the channels depend upon the computational load, which in-turn depends on the type of observation. The output data from the STOKEGEN flow on a single bus, and is picked up by individual nodes using strobes that are generated individually for each node. The data of different frequency channels are multiplexed in time, while the Stokes parameters ( 8-bit each ) are available in parallel, thus forming a 32-bit word stream at a maximum of 16Ms/sec. As shown in figure, the polarimeter module has four major blocks :
1.The input section
2.The Stokes Parameter calculation section
3.The Faraday de-rotation/Gain correction section and
4.Data Distribution Section
1.Input Section:
The input section accepts data from both in-coherent ( IA ) phased-array ( PA ) modes of the G.M.R.T. , the sampling rates remains fixed irrespective of the base-bandwidth selected, thereby resulting in over-sampled data for bandwidths < 16MHz. The FFT spectra will thus have redundant frames, and the number of redundant frames depends on the over-sampling factor. The computing power of the DSP nodes that follow the STOKEGEN can be utilized better by dropping the redundant frames. The frame-rejection logic does this job, and the number of frames rejected between accepted frames is programmable to cover the entire range of G.M.R.T. bandwidths. This rejection is impressed upon both IA and PA samples.
2.Calculation of Stokes Parameters :
This section operates exclusively on PA mode data. The basic equations relating the complex voltage samples to the Stokes parameters are as listed in the following table. The table also lists the number of multiplications and additions required for calculation of each Stokes parameter. For the Nyquist sampling rate , the total computation rate will be 544 million operations per second ( MOPS ) for calculating all four Stokes parameters. This calculation has been implemented as a dedicated pipe-lined logic, occupying about 30% of an 100K-gate FPGA. The first stage of the circuit calculates the desired products and the second stage adds / subtracts these product terms to produce I, Q, U and V. From table, it is clear that depending on the polarization state of antenna and the hybrid ( circular or linear ), the computed results represent different Stokes parameters. This is accommodated so that the results are channelled to the appropriate Stokes parameters based on whether the input channels were linear or circular polarized. The Stokes parameters are quantized to 8 bits each, using a sliding window with round-off logic. The window position can be programmed in software, depending on the strength of the input signals.
3.Faraday de-rotation / Gain Correction:
After calculation of Stokes parameters, the frequency spectrum corresponding to parameters Q + j.U ( the linearly polarized component ) is multiplied with a user-defined phase spectrum Φr + jΦi.This phase spectrum can form the correction factors for Faraday rotation and Parallactic angle. These values are stored in a dual-ported memory can be updates online to track their variations in time. This complex-multiplication is implemented in a set of a pipelined multipliers and adders. This correction is relevant to only the PA mode operation. In the IA path, the samples represent the intensity in each input polarization channel. For data in the IA path, the contents of the same dual-port RAMS are used as gain correction factors for all frequency channels independently for each polarization channel. The IA data from the input section passes through pipelined scalar-multipliers which serve to scale the gains of each channel using the corresponding correction factors from the DPRAMS. After gain correction, the intensity data of two polarization channels are added.
4.Data distribution :
As mentioned above, the PA mode data generates 4 Stokes Parameters in parallel, 8-bit each, every 62.5ns. Every successive clock produces the Stokes parameters of the next frequency channel and a new spectrum starts after every 256 frequency channels. At the same rate and in parallel, the IA mode data produces total intensity samples, 8-bit each. Depending on the observation either the IA or the PA path results are selected for further processing in the DSP nodes. To reduce the transfer rate to DSP nodes in the IA mode operation, 4 frequency channels are packed together to form a 32-bit word. This means an effective transfer rate of about 16Ms/s in PA mode and 4Ms/s in IA mode. To help in cross checking for misses in data transfer, a provision is made to insert a user-defined 32-bit marker pattern in the data sequence, before routing it to the DSP nodes. This marker actually replaces the contents of one frequency channel ( all Stokes parameters in PA mode, total intensity in IA mode ). By default, the marker replaces the first frequency channel and the outputs from STOKEGEN is shared equally by 8 DSP nodes ( i.e. 32 consecutive frequency channels per node ). Separate strobe pulses are supplied to each DSP node, which appear only with the data of channels that correspond to respective nodes. The strobe-pulse sequence is programmed into a PROM. By changing the contents of the PROM, the data can be routed to the nodes. The number of channels supplied to each node can also be changed by altering the strobe sequence, so that a lesser number of nodes can handle all the data for applications with a smaller data rate. The PROM is internal to STKDAS FPGA, and can hence be configured online.