Finite precision arithmetic in Polyphase Filterbank implementations
- Authors: Myburgh, Talon
- Date: 2020
- Subjects: Radio interferometers , Interferometry , Radio telescopes , Gate array circuits , Floating-point arithmetic , Python (Computer program language) , Polyphase Filterbank , Finite precision arithmetic , MeerKAT
- Language: English
- Type: text , Thesis , Masters , MSc
- Identifier: http://hdl.handle.net/10962/146187 , vital:38503
- Description: The MeerKAT is the most sensitive radio telescope in its class, and it is important that systematic effects do not limit the dynamic range of the instrument, preventing this sensitivity from being harnessed for deep integrations. During commissioning, spurious artefacts were noted in the MeerKAT passband and the root cause was attributed to systematic errors in the digital signal path. Finite precision arithmetic used by the Polyphase Filterbank (PFB) was one of the main factors contributing to the spurious responses, together with bugs in the firmware. This thesis describes a software PFB simulator that was built to mimic the MeerKAT PFB and allow investigation into the origin and mitigation of the effects seen on the telescope. This simulator was used to investigate the effects in signal integrity of various rounding techniques, overflow strategies and dual polarisation processing in the PFB. Using the simulator to investigate a number of different signal levels, bit-width and algorithmic scenarios, it gave insight into how the periodic dips occurring in the MeerKAT passband were the result of the implementation using an inappropriate rounding strategy. It further indicated how to select the best strategy for preventing overflow while maintaining high quantization effciency in the FFT. This practice of simulating the design behaviour in the PFB independently of the tools used to design the DSP firmware, is a step towards an end-to-end simulation of the MeerKAT system (or any radio telescope using nite precision digital signal processing systems). This would be useful for design, diagnostics, signal analysis and prototyping of the overall instrument.
- Full Text:
- Date Issued: 2020
- Authors: Myburgh, Talon
- Date: 2020
- Subjects: Radio interferometers , Interferometry , Radio telescopes , Gate array circuits , Floating-point arithmetic , Python (Computer program language) , Polyphase Filterbank , Finite precision arithmetic , MeerKAT
- Language: English
- Type: text , Thesis , Masters , MSc
- Identifier: http://hdl.handle.net/10962/146187 , vital:38503
- Description: The MeerKAT is the most sensitive radio telescope in its class, and it is important that systematic effects do not limit the dynamic range of the instrument, preventing this sensitivity from being harnessed for deep integrations. During commissioning, spurious artefacts were noted in the MeerKAT passband and the root cause was attributed to systematic errors in the digital signal path. Finite precision arithmetic used by the Polyphase Filterbank (PFB) was one of the main factors contributing to the spurious responses, together with bugs in the firmware. This thesis describes a software PFB simulator that was built to mimic the MeerKAT PFB and allow investigation into the origin and mitigation of the effects seen on the telescope. This simulator was used to investigate the effects in signal integrity of various rounding techniques, overflow strategies and dual polarisation processing in the PFB. Using the simulator to investigate a number of different signal levels, bit-width and algorithmic scenarios, it gave insight into how the periodic dips occurring in the MeerKAT passband were the result of the implementation using an inappropriate rounding strategy. It further indicated how to select the best strategy for preventing overflow while maintaining high quantization effciency in the FFT. This practice of simulating the design behaviour in the PFB independently of the tools used to design the DSP firmware, is a step towards an end-to-end simulation of the MeerKAT system (or any radio telescope using nite precision digital signal processing systems). This would be useful for design, diagnostics, signal analysis and prototyping of the overall instrument.
- Full Text:
- Date Issued: 2020
Modelling and investigating primary beam effects of reflector antenna arrays
- Authors: Iheanetu, Kelachukwu
- Date: 2020
- Subjects: Antennas, Reflector , Radio telescopes , Astronomical instruments -- Calibration , Holography , Polynomials , Very large array telescopes -- South Africa , Astronomy -- Data processing , Primary beam effects , Jacobi-Bessel pattern , Cassbeam software , MeerKAT telescope
- Language: English
- Type: Thesis , Doctoral , PhD
- Identifier: http://hdl.handle.net/10962/147425 , vital:38635
- Description: Signals received by a radio telescope are always affected by propagation and instrumental effects. These effects need to be modelled and accounted for during the process of calibration. The primary beam (PB) of the antenna is one major instrumental effect that needs to be accounted for during calibration. Producing accurate models of the radio antenna PB is crucial, and many approaches (like electromagnetic and optical simulations) have been used to model it. The cos³ function, Jacobi-Bessel pattern, characteristic basis function patterns (CBFP) and Cassbeam software (which uses optical ray-tracing with antenna parameters) have also been used to model it. These models capture the basic PB effects. Real-life PB patterns differ from these models due to various subtle effects such as mechanical deformation and effects introduced into the PB due to standing waves that exist in reflector antennas. The actual patterns can be measured via a process called astro-holography (or holography), but this is subject to noise, radio frequency interference, and other measurement errors. In our approach, we use principal component analysis and Zernike polynomials to model the PBs of the Very Large Array (VLA) and the MeerKAT telescopes from their holography measured data. The models have reconstruction errors of less than 5% at a compression factor of approximately 98% for both arrays. We also present steps that can be used to generate accurate beam models for any telescope (independent of its design) based on holography measured data. Analysis of the VLA measured PBs revealed that the graph of the beam sizes (and centre offset positions) have a fast oscillating trend (superimposed on a slow trend) with frequency. This spectral behaviour we termed ripple or characteristic effects. Most existing PB models that are used in calibrating VLA data do not incorporate these direction dependent effects (DDEs). We investigate the impact of using PB models that ignore this DDE in continuum calibration and imaging via simulations. Our experiments show that, although these effects translate into less than 10% errors in source flux recovery, they do lead to 30% reduction in the dynamic range. To prepare data for Hi and radio halo (faint emissions) science analysis requires carrying out foreground subtraction of bright (continuum) sources. We investigate the impact of using beam models that ignore these ripple effects during continuum subtraction. These show that using PB models which completely ignore the ripple effects in continuum subtraction could translate to error of more to 30% in the recovered Hi spectral properties. This implies that science inferences drawn from the results for Hi studies could have errors of the same magnitude.
- Full Text:
- Date Issued: 2020
- Authors: Iheanetu, Kelachukwu
- Date: 2020
- Subjects: Antennas, Reflector , Radio telescopes , Astronomical instruments -- Calibration , Holography , Polynomials , Very large array telescopes -- South Africa , Astronomy -- Data processing , Primary beam effects , Jacobi-Bessel pattern , Cassbeam software , MeerKAT telescope
- Language: English
- Type: Thesis , Doctoral , PhD
- Identifier: http://hdl.handle.net/10962/147425 , vital:38635
- Description: Signals received by a radio telescope are always affected by propagation and instrumental effects. These effects need to be modelled and accounted for during the process of calibration. The primary beam (PB) of the antenna is one major instrumental effect that needs to be accounted for during calibration. Producing accurate models of the radio antenna PB is crucial, and many approaches (like electromagnetic and optical simulations) have been used to model it. The cos³ function, Jacobi-Bessel pattern, characteristic basis function patterns (CBFP) and Cassbeam software (which uses optical ray-tracing with antenna parameters) have also been used to model it. These models capture the basic PB effects. Real-life PB patterns differ from these models due to various subtle effects such as mechanical deformation and effects introduced into the PB due to standing waves that exist in reflector antennas. The actual patterns can be measured via a process called astro-holography (or holography), but this is subject to noise, radio frequency interference, and other measurement errors. In our approach, we use principal component analysis and Zernike polynomials to model the PBs of the Very Large Array (VLA) and the MeerKAT telescopes from their holography measured data. The models have reconstruction errors of less than 5% at a compression factor of approximately 98% for both arrays. We also present steps that can be used to generate accurate beam models for any telescope (independent of its design) based on holography measured data. Analysis of the VLA measured PBs revealed that the graph of the beam sizes (and centre offset positions) have a fast oscillating trend (superimposed on a slow trend) with frequency. This spectral behaviour we termed ripple or characteristic effects. Most existing PB models that are used in calibrating VLA data do not incorporate these direction dependent effects (DDEs). We investigate the impact of using PB models that ignore this DDE in continuum calibration and imaging via simulations. Our experiments show that, although these effects translate into less than 10% errors in source flux recovery, they do lead to 30% reduction in the dynamic range. To prepare data for Hi and radio halo (faint emissions) science analysis requires carrying out foreground subtraction of bright (continuum) sources. We investigate the impact of using beam models that ignore these ripple effects during continuum subtraction. These show that using PB models which completely ignore the ripple effects in continuum subtraction could translate to error of more to 30% in the recovered Hi spectral properties. This implies that science inferences drawn from the results for Hi studies could have errors of the same magnitude.
- Full Text:
- Date Issued: 2020
- «
- ‹
- 1
- ›
- »