Background of the invnention The field of invention relates to a device for compensating the magnetic field of a magnetic resonance (MR) system. More particularly, the invention relates to the measurement of and subsequent compensation for the spatially and temporally varying magnetic fields induced by eddy currents. In doing so, image distortion, signal intensity loss, ghosting, shading, and other artifacts due to eddy currents can be avoided. When a substance such as human tissue is subjected to a uniform magnetic field (polarizing field B0), the individual magnetic moments of the spins in the tissue attempt to align with this polarizing field, but precess about it in random order at their characteristic Larmor frequency. If the substance, or tissue, is subjected to a magnetic field (excitation field 81) which is in the x-y plane and which is near the Larmor frequency, the net aligned magnetic moment, Mz, may be rotated, or "tipped", into the x-y plane to produce a net transverse magnetic moment Mt. A signal is emitted by the excited spins, and after excitation signal 61 is terminated, this signal may be received and processed to form an image. The application of magnetic resonance to imaging, and to many of the techniques of localized spectroscopy, depend upon the use of linear magnetic field gradients to selectively excite particular regions and to encode spatial information within the NMR signal. During the NMR experiments, magnetic field gradient waveforms with particularly chosen temporal variations are used. Any departure from the application of ideal magnetic field gradient waveforms can, therefore, be expected to introduce image distortion, intensity loss, ghosting, and other artifacts. For example, imperfect rephasing of the nuclear spins and an attendant loss of signal occurs if the magnetic field gradients are not constant during selective time reversal pulses (i.e. use of 180° time reversal RF pulses). This effect compounds in later spin echoes of multi-echo (Carr-Purcell-Mieboom-Gill) sequences. In addition, if the gradient field is not zero when it should be (due to residual decay after termination of a gradient pulse), the unintended phase dispersion can result in distorted spectra in chemical shift imaging (CSI) sequences as well as inaccurate spin-spin relaxation time (T2) determination in multi-echo sequences. Those skilled in the art are thus concerned particularly about the accuracy with which time varying magnetic field gradients are produced. Distortion in the production of magnetic field gradients can arise if the gradient fields couple to lossy structures within the polarizing magnet such as its cryostat (if the magnet is of the superconductive design), or the shim coil system, or the RF shield used to decouple the gradient coils from the RF coil. The gradient distortions derive from induction of currents in these ambient structures and from the loss of energy to the shim coils. These induced currents are known as eddy currents. Due to eddy currents, one observes, typically an exponential rise and decay of the magnetic field gradient during and after, respectively, the application of a trapezoid current pulse to the gradient coil. In U.S. Patent No. 4,698,591 entitled "A Method for Magnetic Field Gradient Eddy Current Compensation," a method is disclosed which uses an analog preemphasis filter in the gradient power supply to shape the current applied to the gradient coil in such a way that the eddy current induced gradient field distortions are reduced. The filter includes a number of exponential decay components and adjustable potentiometers which must be set during system calibration. A measurement technique is used prior to system calibration in which the impulse response of the uncorrected magnetic field gradient is measured and the potentiometer settings for the pre-emphasis filter are then calculated. It has been discovered that while such compensation of the linear magnetic field gradients improves performance of MR systems, magnetic field distortions still arise as a result of the application of pulsed linear magnetic field gradients. More specifically, measurements indicate that eddy currents which are induced by magnetic field gradient pulses not only produce an unwanted linear magnetic field gradient, but also cause temporal variations in the spatially homogeneous polarizing magnetic field B0. That is, magnetic field gradient pulses cause spurious changes in the magnitude of the polarizing magnetic field B0. Techniques have been developed to measure and compensate for the eddy current induced B0-field changes, as described in U.S. Patent No. 4,950,994. The magnetic field produced by eddy currents is a complicated phenomenon due to its temporal and spatial dependencies. In order to simplify the problem, the prior corrective methods for eddy current measurement and compensation have assumed that the spatial dependency is limited to only the zeroth (i.e., the homogeneous polarizing magnetic field B0) and first orders (i.e., the linear magnetic field gradients), as illustrated in U.S. Patent Nos. 4,698,591 and 4,950,994. The higher order spatial dependencies (quadratic, cubic, etc) of the eddy current induced magnetic field are left uncompensated, producing residual image artifacts and spectroscopic degradation. Although methods have been developed to address some of the image quality problems, such as geometric distortion as described in U.S. Patent No. 4,591,789, other problems including ghosting, shading intensity reduction, spectrum shifting, and phase errors, still remain. Summary Of The Invention The present invention is an improvement of prior methods used to measure and compensate for the eddy current induced magnetic field distortions. In doing so, the aforementioned image and spectrum quality problems are either eliminated or significantly reduced. More specifically, the present invention includes a method for spatially and temporally resolving variations in the eddy currents that result from the application of a gradient pulse. A series of phase images are produced, and from these the spatial and time-resolved magnetic field produced by the eddy currents is calculated. From this the amplitudes and the time constants of the spatially resolved eddy current components can be calculated and used in subsequent scans to correct for the distortions that are otherwise produced. An objective of the invention is to measure the spatial and temporal variations in eddy currents produced by a gradient pulse. This is accomplished by performing a calibration scan using a calibration pulse sequence. The calibration pulse sequence begins with a test gradient Gt..t, followed by a non-selective RF pulse with an optimal tip angle (i.e., the Ernst angle). The FID induced by the RF pulse is spatially encoded in 1, 2 or 3 dimensions (depending on the geometry of the phantom) using phase-encoding gradients. After spatial encoding, the FID signal continues to precess in the presence of a time-varying magnetic field produced by the eddy currents. Therefore, the temporal behavior of the eddy currents is also encoded in the FID signal. Due to the use of phase-encoding gradients, the time-varying magnetic field is caused by eddy currents arising from both Gtest and the phase-encoding gradients . In order to remove the effects of the latter, as well as the effects of the static B0 field inhomogeneities , the pulse sequence is repeated, but with an opposite test gradient polarity, The two FID signals generated by this method can be denoted as S+(kx, ky, kz, ti) and S.(kx/ ky, kz, tt), where ti represents discrete time points of the FID signal (i=l, 2, ... N), and the other three parameters are the spatial frequencies . A multi-dimensional fast Fourier transformation (FFT) of S+ and S_, with k*, ky and kz being the variables, produces two sets of time-resolved complex images I+(x,y,z,t1) and I.(x,y,z,ti) . The complex images can be readily converted to phase images +(x,y, z , tj.) and =2xy/R2/ dividing α2,ti and ß2,ti by R2 and R2/2, respectively, yields the harmonic coefficients for (x2-y2) and xy. Using this method, some higher order harmonics can also be obtained. After obtaining all the harmonic coefficients at all the time points, the corresponding eddy current amplitudes and time constants can be extracted through curve fitting, and the compensation currents can be applied to the x and y-gradient coils and xy and x2-y2 shim coils, as described earlier. A third test scan uses the same phantom ring 244, but the ring 244 is re-positioned as shown in Fig. 7. More specifically, the phantom ring 244 is translated along the z axis, away from the xy-plane. The exact same test scan used to produce the second calibration data set is then repeated. A third pair of calibration data sets I3) + (Px(x,y) ,ti) and I3,.(Px(x,y),ti) are thus produced and used to calculate the remainder of the quadratic harmonic terms yz and zx. The magnetic field map obtained from I3)+ and I3,- can be expressed as: (Formula Removed) where r0 and θ0 are indicated in Fig. 1. Performing a Fourier transform on S3,ti , the real and imaginary Fourier coefficients for the first Fourier harmonic are found to be: Ignoring the higher order terms beyond quadratic, from Eqs. 10 and 13 one obtains: (Formula Removed) From Eq. 4, it can be seen that 3C12,C and 3D12,ti are equal to the harmonic coefficients for xz and yz, respectively. A curve fitting with respect to time for each harmonic coefficient, as described previously, will give the eddy current amplitudes and time constants. With known eddy current amplitudes and time constants, the spatial eddy current components xz and yz can be compensated by supplying currents to the xz and yz shim coils. The spatially varying eddy currents up to the second order can thus be measured using three separate one dimensional calibration scans, and subsequently compensated by supplying currents to the corresponding B0, three linear gradient, and five second order shim coils. Using the present invention a pure phase-encoding technique is employed to produce a series of images which each represents a true "snap shot," instead of a time-average view of the spatially resolved eddy currents. The time resolution of the eddy current measurement is thus drastically improved over prior methods, and more accurate, higher order compensating currents can be calculated. Claim: 1. A device for compensating the magnetic field of a magnetic resonance (MR) system, the steps comprising computer unit having: - an image processor module - CPU module - Memory module - and a system control unit (122), said computer unit and control unit communicating through a high speed serial link, said control unit including set of modules, including a CPU module, a pulse generator modules, and a module configured for a) acquiring a first calibration data set using a pulse sequence by a pulse generator module (121) comprises the steps of: applying a test gradient pulse (202) of one polarity: applying an RF excitation pulse to produce transverse magnetization in a region of interest; applying a phase encoding gradient pulse; and acquiring a nuclear magnetic resonance (NMR) signal over a time period (T) following the application of the test gradient pulse and sampling it at times t: wherein the pulse sequence is repeated a plurality of times and the phase encoding gradient pulse is stepped through preset values; b) acquiring a second calibration data set as recited in step a), except the test gradient pulse that is applied has the opposite polarity; c) Fourier transforming each of the two calibration data sets to produce two sets of spatially and temporally resolved phase images; at processor block (222) d) subtracting the second set of phase images from corresponding phase images in the first set phase images to form a phase-difference image set; at a processor block (226) e) calculating eddy current compensating values based on the phase difference images; and f) applying the compensation values to coils on the MR system during subsequent scans.