As the signal is passed through the atmosphere, it is degraded in multiple ways. The degradation at each step is modeled, including the ionosphere. These are computed on the base station through modeling, and the corrections are sent back to the satellite. Ionospheric corrections are then passed to the receivers (rovers) either through the satellites or through internet (via IP connection). Sometimes IP connection cannot be used, or is avoided. Information sent through the satellites travels at the rate of 1200 bps, with the price tag of $250-500/bps. If this information is compressed (taking smaller bandwidth), some of that bandwidth expense can be saved.
Technical Problem description: e.g. How do you define the target variable?
The major issue with the current approach is that the client currently has a complex model that computes the ionospheric corrections for each individual satellite/base station pair and all corrections are then sent to all receivers within satellite range. This is highly inefficient, we must engineer a model that will be capable of reducing either the amount of data that needs to be transmitted, or the frequency of transmission.
Baseline Approach: How can we estimate a baseline performance?
The performance can be evaluated in terms of file size savings per unit time (on the correction files that are to be passed to the receiver). Error due to compression has to be within 1cm for the final position. Compression/decompression time limit has to be around 2-7 seconds for each procedure.
From the base station location (lat/long) we can compute the VTEC(vertical total electron count) of the ionosphere, which is proportional to the amount of ionospheric interference at each satellite location. We will first attempt to model the VTEC concentration as an univariate time series. By plotting and computing the auto-correlation and partial auto-correlation plot we can determine whether or not auto-regressive features are valuable and whether to model the ionosphere as a time series or stationary model.
Once this is determined, we will use a neural network to predict the VTEC for any given satellite location, angle etc…, and map the VTEC value back to TEC. This model can be uploaded onto the OM7700 chip in order to predict the VTEC values and compute the corrections instead of sending the correction data via L-band. The advantage here is that we can send a single model which will make predictions for all satellites instead of a set of corrections per satellite, also if we manage to capture time dependent relationships we can utilize that to predict the ionosphere corrections for a particular time window without having to resend the model.
