Initial modelling approach

The initial task facing the analysis group is to decide which of various approaches might be best - our resources are likely to be too thin, unless a large amount of people join the analysis group, to take on too many approaches in parallel.  The first task will therefore be more down to the experts that we hope to attract to the group, who will map out the best way forward.  

This may be one of several routes, that includes multivariable analysis, outlined below, and possibly techniques such as neural networks.  However, there could be some more direct approaches, for example mapping the data onto existing analytic approaches to describing propagation.  

Multivariable analysis

This approach is based upon the concept of  taking each of the main variables that we have been recording in the experiment, and seeing the extent to which they can be combined in a systematic way such that the performance, or quality, of an arbitrary path across the UK can be modelled.   Once the model has been produced and validated using our 5MHz data it should be capable of being used in the future to predict the quality of paths across the UK.  In effect, it will be a finely tuned propagation prediction utility that is specific to inter-UK paths, many with NVIS characteristics.

The variable that we decided to record in the experiment were as follows: Time & date, location, various parameters of signal and noise, channel, mode, power, aerial type (5 different categories), aerial height, polarisation and orientation.   In addition from other sources we have tables (with relational capabilities to our main log data tables) that provide a variety of measured, and in some cases averaged, propagation indices.

Modelling the variables

This is an area of work where the Amateur can play a real part with his or her knowledge of ionospheric physics and the system aspects of radio links.  This is an area of work where we will though also need some specialist mathematical, statistical and numerical analytical input, either from within the Amateur community and/or from academia.

We know from prior knowledge quite a lot about some of these variables.  For example we know that time and date will follow some pattern that is related to solar illumination of the refracting layers in the ionosphere, which are also at the mid-points of our normal "single hop" propagation path.  The starting point for modelling these variables may well be to start using polynomials which contain cosine and sine functions, so that the basic shape of the diurnal and other periodicities, such as seasonal and the nominal 11 year sunspot cycle can be modelled with the minimum number of terms in the polynomial. 

We might choose to simplify our use of the "location" variable, by merely expressing either the path length between two stations, or perhaps defining just a small range of take-off/arrival angles.  Perhaps later, we may choose to see if we can differentiate between the more northerly and southerly locations, or perhaps paths that are N/S or E/W, etc.  The point of trying to simplify the number of variables that is in effect removed one gets a great deal more "accuracy" (or what the statisticians call significance) from one's modelling.  

Our decision to use and then tighten the way that SINPO was defined should help us gain some insight to the signal and noise environment.  This should further enable us to be able to show some relationship between the automatically recorded data from the beacon monitoring, manual beacon monitoring (via recording of 6dB steps relative to the noise floor) and the more subjective SINPO reports of individual QSOs.  Quite how this is best modelled is open to more discussion and no doubt hard work, but the starting point may be some first or second order polynomials (for the more mathematically inclined!).  

We may choose to do some separate statistical work to see if there is much difference in terms of propagation performance re Channel though the small frequency difference between channels would not suggest much change is likely. However, the effects may be see more in terms of noise or interference, though it may be difficult to separate these different causes and they may not have too much bearing on the future use of the model.

Modelling the propagation indices will require significant thought.  Whilst some of the indices have values that are somewhat random in nature, others follow the solar sunspot cycle albeit with perturbations.  This will be an area where we will require those more versed in ionospheric propagation to influence the modelling,

The remaining variables are more to do with overall S/N taken from mode efficiency, transmit power and overall aerial efficiency (relative to the propagation path).  The modelling of these parameters, like SINPO may need a more generalised approach as there may not be any underlying physics relating to the variable on which to make the modelling more specific.

Combining the variables

This is an area of work in which we will need even more specialist input and numerical analysis skill than in the initial modelling of the variables.  Multivariable analysis is a well developed discipline and one which we will need to use in the main part of this analysis project.  However, somewhat similar work to the problem that we are attempting to solve has been published in professional journals in recent years, so we can be optimistic of the outcome.  The main issue is likely to be the amount and quality of the data that we have collected.  The former or course is somewhat dependent upon whether we can continue with the experiment after the current period of access to the 5MHz channels.  The more data we get also helps to smooth out and hide areas of poor quality in the data.