by Robert Smith, Stuart Glynn and Liam Devlin, PRFI
The growth in broadband 5G and satellite communications applications has created an increased demand for high performance, low-cost mmWave Power Amplifier (PA) components. The datasheets and design briefs for these components often use traditional microwave PA performance metrics such as output power at 1dB gain compression (P-1dB) and third order intercept point (IP3). While these metrics can provide useful indications of the likely performance within a system, there are other key parameters that are more directly relevant to the system designer.
Modern communications systems use complex digitally modulated waveforms. As the modulated signals pass through the system, they suffer amplitude and phase distortion, and the output PA is normally a significant contributor to this distortion. For the system designer, it is the level of distortion of the complex modulated waveform that the PA causes which is of most interest. Performance metrics such as Error Vector Magnitude (EVM) and Adjacent Channel Leakage Ratio (ACLR) are a better indication of how well a PA will operate within the system than the more conventional metrics found on most datasheets. Fortunately, the availability of accurate large signal transistor models and appropriate simulation tools allow these system level parameters to be accurately simulated.
Figure 1 shows a photograph of a plastic packaged 28 GHz PA MMIC designed by PRFI for a 5G application (an image of the internal die is shown above the evaluation PCB for the plastic packaged component). The PA covers 26.5 to 29.5 GHz so it supports both the n257 and n261 mmWave 5G bands. It was designed on a commercially available 0.15µm gate length E-mode PHEMT process and includes a power down circuit and an on-chip temperature compensated RF output power detector. The IC was assembled into a low-cost plastic overmolded 4mm x 4mm QFN package and all measurement data shown here was for the packaged component on the evaluation PCB of Figure 1.
The PA was designed against a specification based on conventional amplifier metrics (gain, return losses, IP3, etc.). In order to confirm its suitability for the intended 5G application, it included target specifications for performance when operating backed off to the point where third order intermodulation products (IMD3 products) were at -35dBc. This conventional performance metric was determined as the operating point where the amplifier would add an acceptable level of distortion to the modulated 5G spectrum.
The measured s-parameters of the PA are plotted against the simulated s-parameters in Figure 2. The simulated performance included a full EM simulation of the IC and the good agreement that this produced is evident. Small signal gain (S21) is around 20dB across the 26.5 to 29.5 GHz band with an input return loss (S11) of better than 16dB across the band. The use of an E-mode process meant that the amplifier required no negative supply voltages, and the quiescent bias current was 180mA from a +4V supply.
Initial large signal evaluation of the PA was against conventional performance metrics and demonstrated a typical output P-1dB of +25dBm with an associated Power Added Efficiency (PAE) of around 28%. The measured OIP3 of a typical PA was around +32.5dBm across the 5G band and showed very little variation with tone power over a 10dB dynamic range. The two tone IP3 measurements were made with a tone spacing of 100 MHz to reflect the wide operating bandwidths of 5G modulated signals. At the specified IMD3 level of -35dBc, the RF output power was +18dBm.
The performance of the amplifier met the IMD3 specifications, however the best way to assess the PA’s performance in a 5G system is to evaluate it using real wideband 5G signals. Testing was subsequently undertaken using modulated signals including an OFDM waveform of 400 MHz bandwidth with 120 KHz sub-carrier spacing using 64 QAM modulation. This waveform has a high peak to average ratio of 11.71dB. The right-hand graph in Figure 3 shows the measured Error Vector Magnitude (EVM) against input power at 28 GHz when the PA is transmitting this modulated signal. It can be seen that good linearity is evident and at the intended operating point of 18dBm average output power, the input power is -3dBm and the EVM is less than 4%. The corresponding PAE at this operating point is 8%.
Although EVM was not simulated as part of the original design process, the availability of suitable CAD software allowed subsequent simulation with the same modulated signal as used in the EVM measurement set-up. The EVM simulation was performed using the Envelope Simulation tool from Keysight and the resulting simulated performance is plotted in the left-hand graph of Figure 3. Very good agreement between measured and simulated EVM is evident. This demonstrates that with the right simulation tools and accurate component modelling, it is possible to accurately predict the performance of 5G system metrics such as EVM and ACLR.
It is expected that mmWave PA design specifications will always include traditional performance metrics such as gain, return losses and RF output power requirements. The benefits of being able to accurately simulate the amplifier’s impact on modulated signals in terms of system level performance parameters is that it avoids the need to rely solely on translating modulation dependent system level specifications into conventional amplifier performance metrics. It also reduces the risk of amplifier performance falling short when evaluated in the intended system.