# LinearShimDB: A subset of the NMR magnet shimming database ShimDB

LinearShimDB is a subset of the NMR magnet shimming database [ShimDB](https://github.com/mobecks/ShimDB) and contains over 9000 instances. Data is acquired on a Spinsolve 80 Carbon spectrometer (Magritek GmbH, Aachen, Germany, www.magritek.com) using a water solution with CuSO4 (5mmol/L). 

LinearShimDB is part of "Deep Regression with Ensembles enables Fast, First-Order Shimming in low-field NMR" by M. Becker et al. [1].

The acquisition procedure was as follows. The manufacturer's automated shimming technique, based on the downhill simplex method, was used to obtain a reference spectrum of decent quality. Then, all shim values except the three linear shims X, Y and Z were set to zero. The resulting spectrum and corresponding shim settings were used as the reference values. The database parameters were obtained by relative offsets from the reference shim values in a range R with stepsize s, in a grid-like manner. For each combination, the raw FID, acquisition parameters, and the shim values were stored. 

| Topic                  | Parameter        | Value     |
|------------------------|------------------|-----------|
| Characteristics        | Nr. Spectra      | 9261      |
|                        | Shim range R     | +/- 10000 |
|                        | Step size s      | 1000      |
|                        | Shims            | X,Y,Z     |
| Acquisition parameters | Nucleus          | 1H        |
|                        | Bandwidth        | 20kHz     |
|                        | Points           | 32768     |
|                        | Dwell time       | 50us      |
|                        | Repetition time  | 2000ms    |
|                        | Filter           | -        |
|                        | Phase correction | phi0 |      
       
**We strongly encourage researchers to extend ShimDB with their own subsets to stimulate developments. We offer to include raw data or links to your publications into ShimDB.**


## Files format

Each folder in LinearShimDB contains the following files: 
- data.1d -> the raw FID.
- shims.par -> Shim values, where only linear shims are non-zero.
- acqu.par -> Acquisition parameters.
- proc.par -> Processing parameters.

The LinearShimDB root folder also contains the reference starting shims (ReferenceShims.par).


## Data loading

We deliver a python script ```utils_IO.py``` alongside [ShimDB](https://github.com/mobecks/ShimDB) to easily load the database into numpy array structure using the [nmrglue packages](https://github.com/jjhelmus/nmrglue)[2].

The following python libraries and packages are required: os, numpy, glob, nmrglue (>= v0.9.dev0)


# References

[1] M. Becker, M. Jouda, A. Kolchinskaya, J. G. Korvink, Deep regression with ensembles enables fast, first-order shimming in low-field NMR, Journal of Magnetic Resonance 2022, 107151, ISSN 1090-7807, https://doi.org/10.1016/j.jmr.2022.107151
[2] J.J. Helmus, C.P. Jaroniec, Nmrglue: An open source Python package for the analysis of multidimensional NMR data, J. Biomol. NMR 2013, 55, 355-367, http://dx.doi.org/10.1007/s10858-013-9718-x

LinearShimDB: A subset of the NMR magnet shimming database ShimDB
LinearShimDB is a subset of the NMR magnet shimming database ShimDB and contains over 9000 instances. Data is acquired on a Spinsolve 80 Carbon spectrometer (Magritek GmbH, Aachen, Germany, www.magritek.com) using a water solution with CuSO4 (5mmol/L). 
LinearShimDB is part of "Deep Regression with Ensembles enables Fast, First-Order Shimming in low-field NMR" by M. Becker et al. [1].
The acquisition procedure was as follows. The manufacturer's automated shimming technique, based on the downhill simplex method, was used to obtain a reference spectrum of decent quality. Then, all shim values except the three linear shims X, Y and Z were set to zero. The resulting spectrum and corresponding shim settings were used as the reference values. The database parameters were obtained by relative offsets from the reference shim values in a range R with stepsize s, in a grid-like manner. For each combination, the raw FID, acquisition parameters, and the shim values were stored. 



Topic
Parameter
Value




Characteristics
Nr. Spectra
9261



Shim range R
+/- 10000



Step size s
1000



Shims
X,Y,Z


Acquisition parameters
Nucleus
1H



Bandwidth
20kHz



Points
32768



Dwell time
50us



Repetition time
2000ms



Filter
-



Phase correction
phi0



We strongly encourage researchers to extend ShimDB with their own subsets to stimulate developments. We offer to include raw data or links to your publications into ShimDB.
Files format
Each folder in LinearShimDB contains the following files: 

data.1d -> the raw FID.
shims.par -> Shim values, where only linear shims are non-zero.
acqu.par -> Acquisition parameters.
proc.par -> Processing parameters.

The LinearShimDB root folder also contains the reference starting shims (ReferenceShims.par).
Data loading
We deliver a python script utils_IO.py alongside ShimDB to easily load the database into numpy array structure using the nmrglue packages[2].
The following python libraries and packages are required: os, numpy, glob, nmrglue (>= v0.9.dev0)
References
[1] M. Becker, M. Jouda, A. Kolchinskaya, J. G. Korvink, Deep regression with ensembles enables fast, first-order shimming in low-field NMR, Journal of Magnetic Resonance 2022, 107151, ISSN 1090-7807, https://doi.org/10.1016/j.jmr.2022.107151
[2] J.J. Helmus, C.P. Jaroniec, Nmrglue: An open source Python package for the analysis of multidimensional NMR data, J. Biomol. NMR 2013, 55, 355-367, http://dx.doi.org/10.1007/s10858-013-9718-x