This resource is designed to help you optimise and refine the DPD implementation for improved performance your application. Whether you're new to DPD or looking to fine-tune advanced parameters, this guide provides the insights and tools you need to achieve better linearization and efficiency in your signal chain. Digital Pre-Distortion is a technique used to correct non-linearities in power amplifiers, enabling higher efficiency and maintaining signal integrity. By adjusting key parameters such as delays, time filter coefficients, and normalization thresholds, you can significantly enhance system performance and stability. Below we will go through some pre-calibration and post-calibration parameters to help get the most out of the DPD. Pre-calibration Parameters Enable/Disable LUT size and pre-LUT scale Model Tap polynomial Terms Enable and disabling the DPD can be done from the tracking calibrations tab in the TES. This allows the user to stop the updating of the calibration co-efficients when needed. If the DPD is stable and the enviornmental conditions are stable then there might be no need for iterating the co-efficients and the option is there to pause the calculation. The coefficients currently in use will be kept in this scenario until the DPD is enabled again and the calculation engine calculates new co-efficients. LUT size has an effect on the DPD coefficients. A smaller LUT will have slightly coarser mapping between the input and the coefficient output. It will however save on memory and potentially be slightly faster to calculate coefficients. Model Tap Polynomial Terms can be turned on or off and tested out to see what good stable performance for the signal under test. The values on each of the terms is calculated by the DPD engine and this array in the TES aloows the user to chose what taps are used in the calculation or not. Post Calibration Parameters Number of samples captured Capture Delay Time filter co-efficient/LUT switching Normalisation Thresholds Activation conditions Additional power scale The number of samples captured can be up to 4096 consecutive samples. In the case where the frame might be smaller than the maximum amount of samples you can reduce the cature length. In the other event where th frame is veryy long (seconds) or in FDD mode when its constantly on the DPD will capture every 2 seconds. The Capture delay can be used to delay the start of the DPD capture. In scenario 1 the gaurd time might be taken into account with null data so its possible to push the capture further into the frame to ensure you only get valid data points. The Time filter coefficient allows te DPD to gradually make changes over time. Instead of taking in a new capture and updating the coefficients based solely off the incoming data the time coefficient aloows a blend from the current DPD coefficients to the next ones by adding them and multiplying by the time coefficient. This will mean there is no major jump from one iteration to the next and the DPD will take longer to settle as a result. Normalisation thresholds are important to declare the linear region to normalise the data to. This should be above the noise and below the compression region. The polynomial function creates multiple copies of the the signal based on the order of the non-linearity and memory depth. Additional Power scale is used to to reduce the corelation of the signal to prevent overfitting. The default value is 4. Peak detection activation is another control that will allow the system to make decision on whether to update the DPD coefficients or not. Users can set the threshold and make a decision based on how many times the peak of the signal has crossed that threshold. For example you can set the threshold at -10dBFS and allow the DPD update if the number of points in a capture above the threshold is above 100. If these conditions are not met then the data is discarded and the previous coefficients remain in place. Crest factor reduction needs to be applied to the signal before transmitting, there are no internal CFR blocks in the ADRV9002. Signals like LTE will have higher peak to average power ratios (PAPR). This can cause inefficient use of the LUTs since high peaks take up more resources when compared to lower power samples. There is a trade off needed here to keep the signal integrity but also reduce the peaks to increase the efficiency.

Digital Pre-Distortion

Introduction to DPD

This tutorial will cover a brief intro on Digital Pre-Distortion (DPD), why its used and how it works. It will also cover how to go about setting up a DPD measurement, including running calibrations for losses in equipment and give two examples using different Power Amplifiers (PA). The data shown in this tutorial is from real in lab measurements.
This tutorial is for anyone looking to implement or evaluate the performance of the ADRV9002 or ADRV9005 DPD feature.
To follow along with the practical you will need an ADRV9002 evaluation platform, Transceiver Evaluation Software from the latest SDK, PA and accompanying attenuators and splitters based on your requirements (more on how to chose these below).

Digital Pre-Distortion

DPD is a feature that uses a real-time feedback loop to pre-distort data being transmitted through a power amplifier (PA). It's used in transceiver applications to generate a more robust linear signal for transmitting data, to keep within ACPR specifications for RF bands and to utilise the PA in its most efficient state to maximise power useage. The PA is most efficient in its compression region, as shown in Figure 1, where the signal gets non-linearitiese introduced as a result. DPD works by taking in the output of the PA, calculating the non-linearity and applying an opposite distortion to the signal before it is transmitted and reaches the PA. The sum of this pre-distortion and the PA distortion should output a perfectly linear signal. This allows the user to get the best performance out of the chosen PA, however each PA will have different performance even with DPD and will need to be fine-tuned.

For more in-depth understanding and debugging refer to the User Guide for all up-to-date documentation on the DPD and TES functionality.

Figure 1. PA Output

FAQ

What is the maximum useable bandwidth for the DPD?
The maximum bandwidth for DPD on the ADRV9002 is 20MHz. This is based off 1/5th of the maximum internal Tx observation bandwidth of the DPD block, which is 100MHz. Using this much margin allows the DPD engine to correct some of the odd harmonics that will have the biggest impact on stability.

Setup for DPD

Measurements

The DPD engine requires both Tx and Rx signals to calculate the coefficients. The coefficients will be applied in the Tx signal (pre-distort). Shown in

Figure 2. DPD HW Setup

This setup means that for the user’s implementation, one of the Rx channels will need to be permanently used as the DPD observation path. In the ADRV9002, this means the user can implement one of the following:

1T1R TDD with DPD
1T2R TDD with DPD
2T1R TDD with DPD
2T2R TDD with DPD
1T1R FDD with DPD

Figuring out what your hardware setup will include is the next step. To do this we would recommend starting with the frequency band your application is targeting, from there you will need to understand the link budget and how much power output will be needed from the transmit system. From this point you can start to look for a suitable PA for the system. When you have found a PA that will operate in the freq of interest and provide enough power you can determine the rest of the DPD feedback loop. Using the input power required to push the PA into the compression region you can determine if an addition gain block is needed before the PA as the output from the EVB will be between 5 and 7dBm max. The ouput power of the PA will determin how much attenuation is needed to bring the loopback signal down to the required -18dBm to the input of the ADRV9002. This will also help identify the power rating on the attenuators needed.

At this stage you will have a PA, attenuators, gain block and coupler/splitter chosen for the feedback loop.

Example 1

Equipment List

ADRV9002 evaluation platforrm
PC with latest TES
Spectrum Analyser
Signal Generator
Power sensor
PA / Gain block
RF cables
Attenuators
Splitter/Coupler

Calibration

Calibrating each piece of the feedback loop is very important to get the most accurate picture of what is happening in the loop and will help evaluate and stabalise the DPD. Calibration is pretty simple but time consuming as it involves using a signal generator and a power sensor to input a signal of good quality and know freq and amplitude into a cable or coupler and sensing the power on the other end of the part under test. The difference in power level through the part is the loss and should be noted. Running this calibraion at the freq uency of interest is important as the response of each part can change against frequency.

You can use the losses measured to fine tune the attenuation levels in the loopback.

Hardware setup

Shown in Figure 3 is a block diagram of the setup using the ADRV9001 evaluation board (EVB) Tx1 is the transmit channel. The Tx signal must first go through a filter to get rid of harmonics that may make the DPD unstable. The Tx signal is then passed through a simple Minicircuits gain block to increase the signal power level to be able to push the PA into saturation. The gain block is a fixed 40dB so we have the flexibililty on the ADRV9002 side to use the attneuation to set the initial power level for the loopback. The PA outputs around 40dBm so the attneuators next to the PA must be rated to be able to take the 10W of power. In this loopback there is a 20W 20dB attneuator followed by a 10W 10dB and then 2W 20dB to bring the power level down to -10dBm going into the coupler. The coupler also has its own losses that we have claibrated earlier. The cables also have a loss and the sum of these leaves us with -18dBm going into the Rx port on the EVB. There is a little bit of torrerance here, the power level to the Rx1B input should be -18dBm +-5dB for the DPD to operate correctly. This level is not user configurable.

The user needs to be aware of the kind of data being transmitted, for example if the data has a high PAR then the Tx data will need to be backed off enough so as not to saturate the DACs when the signal goes through the peak expansion (pre distortion). There is no crst factor reduction block in the ADRV9002, the user should be aware of this and make provisions to apply crest factor to all the signals being supplied to the ADRV9002 for transmitting. The importnace of this and other tuning methods will be discussed in the advanced DPD tutorial.

Figure 3. Simple Example DPD Block Diagram

TES setup

With this simple setup the user can then use the TES to enable the DPD. For more details on any of the steps below please see the User Guide section on Digital Pre-Distortion and ADRV9001 Evaluation System.

After successfully connecting to the FPGA, in the Configure tab select an LTE TDD setup.

Figure 4. Device Configuration

The external path delay will need to be calibrated. This can be done in the with the Python interface by loading the ‘External_Delay_Measurement.py’ script. This will return the value in ps for the ‘Path Delay’ in the next step. This file is located in the install folder for the TES C:\Program Files\Analog Devices\ADRV9001 Transceiver Evaluation Software\Examples\Python

Figure 5. Python Script Window

In the Board Configuration tab select the external loopback on the channel that will be performing DPD. Input the peak power and the calibrated path delay.

Figure 6. Board Configuration

In the Digital Pre-Distortion section of the Advanced Features tab the user can now enable the DPD on the channel of choice. There are options to select the Look Up Table size (larger number of entries provides more granularity)

The Pre-LUT Scale can be used to get better DPD performance if the input signal to the DPD block is too small. The default value is 2 and be changed from 0 to 3.75 in 0.25 steps.

The polynomial terms can be set to Default or when further optimising the user can custom tune the Taps for a better response. For now, we use the default setting.

Figure 7. Advanced Features (DPD)

In the Digital Pre-Distortion tab, the channel enabled in the Advanced features tab above will also appear enabled here. This tab gives the user the option to increase or decrease the number of samples of real data used to calculate the coefficients. There is an additional power scale that can be used for small input signals.

DPD Activation can be used to prevent the DPD from going unstable if the signal level going to the DPD block drops unexpectedly. For now, these settings can be left as the default values.

Figure 8. Digital Pre-Distortion Tab

The part can now be Programmed. Enabling the DPD will require a DPD tracking calibration which looks for a feedback signal. If the hardware setup is not set up correctly the device will fail to program.
1. Note: In TES version 0.21.0 the DPD updates the coefficients at the start of a frame. If you are using the Automated TDD function the frame length is known and will occur frequently compared to operating in normal mode. In normal mode you will need to Stop and Start the Tx transmitting to trigger a frame and update the DPD coefficients.
2. Note: Make sure that the Transmit file being used has no discontinuities between the start and end of the file. The file is being sent in a loop and discontinuities will cause some unwanted spurious that will make destabilise the DPD performance.

Measurement Results

Doing the same measurement with and without DPD is the best way to show the impact. To run without DPD you will need to reset the part, disable DPD and program the part again. The results from this measurement are shown in Figure 9. The yellow trace is the result before the DPD is used and the pink trace is with DPD enabled. This shows a 16dB improvement however all setups with differing PAs will produce different results. Figure 9 is for illustrative purposes only.

Figure 9. Signal Analyser Measurement Example Measurement

The measurements below were taken using the ADRV9002 customer evaluation board controlled by the Transceiver Evaluation Software. The measurements show the results from two different PAs with DPD and without.

Measurement 1 - Tetra1

PA: Sky65160-11

ADRV9002 Profile:

Tetra1, 144Ksps
Internal LO
Tx Continuous On

Transmit Waveform:

TETRA1_sample_rate_144K_bw_25K
- This can be found in \ADRV9001 Transceiver Evaluation Software\Examples\Tx Data\
PAR is about 3dB

Figure 10: Measurement 1 - Tetra1

Measurement 2 - LTE10M

PA: Sky66297-11

ADRV9002 Profile:

TDD LTE10, 15.36Msps
Internal LO
Tx Continuous On

Transmit Waveform:

CFR_sample_rate_15p36M_bw_10M
- This can be found in \ADRV9001 Transceiver Evaluation Software\Examples\Tx Data\
PAR is about 9dB

Figure 11: Measurement 2 - LTE10M

Measurement 3 - LTE20M

PA: Sky66297-11

ADRV9002 Profile:

TDD LTE20, 15.36Msps
Internal LO
Tx Continuous On

Transmit Waveform:

CFR_sample_rate_15p36M_bw_10M
- This can be found in \ADRV9001 Transceiver Evaluation Software\Examples\Tx Data\
PAR is about 9dB

Figure 12: Measurement 3 - LTE20M

MATLAB DPD Analysis Tool

The MATLAB tool is aimed at analyzing the captured data from dpd_capture.py. This is an IronPython script that can be run from the TES to capture Tx and Rx data for analysis. This tool will help check signal integrity, signal alignment, PA compression level and at last the fine tuning of DPD.

Figure 13: MATLAB DPD analysis tool

This DPD analysis tool takes in a .csv file that is gererated in the TES with the dpd_capture.py script. This is run from the Python editor in the View tab.

Figure 14: Python Editor

For more in depth understanding of the DPD features and setup have a look at the whitepaper on The Complete Guide to Troubleshoot and Fine Tune Digital Predistortion

Running this will require a MATLAB runtime engine and this is included in the install package for the DPD Analyser.

The ADRV9002_DPD_Analyser can be downloaded here: ADRV9002_DPD_Analyser.zip

The dpd_capture.py file can be downloaded here: dpd_capture.zip

DPD Power Consumption

The ADRV9002's DPD power consumption, while seemingly dominated by the Observation Receiver (ORx), is actually primarily attributed to the DPD Actuator in average power calculations. This is because the ORx, despite consuming around 225mW for Tetra and 281mW for LTE 20 1T/1R during operation, only runs periodically for data capture (approximately every 2 seconds). The capture duration itself is short, ranging from tens of microseconds to several milliseconds depending on the profile: around 7.1ms for Tetra and 23µs for LTE 20.

Therefore, the average ORx power consumption is significantly lower: around 0.8mW for Tetra and a mere 0.003mW for LTE 20 1T1R. This is calculated by multiplying the ORx power consumption with the ratio of data capture time to the period. Consequently, the ADRV9002's average DPD power consumption is mainly determined by the always-on Actuator, resulting in less than 3mW for Tetra, less than 18mW for LTE 20 1T, and less than 36mW for LTE 20 2T. More details on DPD power consumption are provided here:

PDF

DPD Performance Reports

The ADRV9001 employs DPD system similar to that of the AD9371 and AD9375 transceivers. The published performance reports for these transceivers with various PAs are listed below. Comparable DPD performance can be expected with the ADRV9001 using similar PAs, provided the instantaneous bandwidth (iBW) requirements are met.

Tags: transmit ad9371 Transceiver Evaluation Software ADRV9002 ADRV9001 Gain Blocks dpd ad9375 rf amplifiers Wideband Transceiver IC adl5611 RF Integrated Transceivers Show More