Many-coil transmit arrays are desirable in parallel transmission, since with many coils multidimensional pulses can be shortened, more uniform RF shims can be produced, and SAR can be more effectively controlled. However, the high cost and the large physical footprint and cabling requirements of the corresponding power amplifiers required to drive many-coil arrays has limited the number of transmit coils/channels used in practice, and most ultra-high field MR scanners in use today have only 2 or 8 transmit channels. To overcome this limitation, array-compressed parallel transmission (acpTx) has been recently proposed [1,2] in which a large number of coils ($$$N_{coils}$$$) is connected to a small number of channels ($$$N_{chan}$$$) via a hardware compression matrix of tunable attenuators and phase shifters. While this concept has the potential to match the performance of a many-coil system with a small number of channels, it has not yet been experimentally demonstrated. In this study an acpTx hardware system was built and used to demonstrate acpTx using two transmit channels/amplifiers to drive an 8-coil array at 7T.

AcpTx Hardware System: An array compression matrix was built to implement the 1 channel-to-4 coils acpTx system architecture illustrated in Figure 1 [2] at 7 Tesla. The matrix takes two transmit channel inputs, and splits them each to four transmit coils using three Wilkinson power splitters (Fig. 2C). Each coil’s signal is then attenuated using an adjustable attenuator (Figs. 2A,B) comprising a power splitter connected to a power combiner using adjustable transmission line lengths; the relative length of the two lines determines the attenuation. Finally each coils’ signal is phase shifted by another length of transmission line. A custom 8-coil head array was also built (Fig. 2D), which incorporated magnetic wall decoupling [4]. The electromagnetic coupling between any two loops in that array was better than -15 dB.

AcpTx Pulse Design: Accelerated acpTx spiral-in pulse waveforms (R=3.3 acceleration factor) and array compression weights for the hardware matrix were jointly designed using the method in [1,3]:

$$\textrm{minimize } \phi(B); \textrm{s.t. rank}(B(\Theta_i)) = 1, i = 1,2. \textrm{ (Eqn 1)}$$ where $$$\phi$$$ is the standard spiral pulse design cost function [3], $$$B$$$ is the RF pulse matrix for all $$$N_{coils}$$$ transmit coils and pulse time points, and the $$$\Theta_i$$$ are coil-to-channel assignment vectors. In this study, $$$\Theta_i$$$ were chosen to split the array in two circumferentially, mapping four consecutive loops to each input channel.

Experiment: Experiments were performed on a Philips 7T Achieva scanner (Philips Healthcare, Cleveland, Ohio, USA). Prior to pulse design, the $$$B_1^+$$$ maps of all 8 coils were measured using DREAM [5]. The acpTx pulse design was run (to solve Eqn 1) and the resulting array compression weights were normalized in amplitude and phase by the weight with the largest amplitude in each combination, and were then implemented using the attenuators and phase shifters in the hardware compression matrix. The pulses’ excitation patterns were imaged using a 3D gradient echo sequence (TE/TR = 2/1000 ms, flip angle = 5 degrees, FOV = 20x20x20 cm, voxel size = 3x3x5 mm), and 32 channel-receive (with 3D SENSE acceleration of 2 in AP and FH directions). For comparison, two-channel transmit pulses using the same spiral trajectory were also designed for the I and Q ports of a circularly-polarized (CP) combination of the coils.

Figure 3 shows predicted and measured $$$|B_1^+|$$$ maps for an arbitrary two-channel compression with two coils’ signals attenuated (by -3 dB & -5 dB) and with three other coils’ signals phase-shifted (-40, -60, and -100 degrees). The maps agree closely, indicating that the hardware array compression matrix functions as predicted. The acpTx spiral pulse design generated weights with nearly uniform amplitudes, so the matrix only needed to apply phase shifts in that unique case (maps not shown); however, in general both attenuations and phase shifts are required. Figure 4 shows the predicted and measured spiral excitation patterns. The two-channel CP-mode excitation had higher errors in both the predicted and measured patterns (some of which are indicated by the green arrows), and there is good agreement between the predicted and measured patterns in both cases.In this work, an array-compressed parallel transmit hardware system was developed and validated. The central component of the system was a tunable 2 channel-to-8 coil compression matrix. Its ability to apply array compression weights resulting from array-compressed parallel transmit pulse designs was validated through $$$B_1^+$$$ mapping and accelerated spiral excitation experiments, and in the latter experiments it was shown that array-compressed pulses achieved a more accurate excitation pattern than two-channel CP-mode pulses.