LinkSwitch-TN2Q reference desing (RDR-707Q) roadtest

Table of contents

RoadTest: Review the LinkSwitch-TN2Q Non-Isolated Buck Power Supply RDK-707Q

Author: JWx

Creation date: 20 Nov 2024

Evaluation Type: Evaluation Boards

Did you receive all parts the manufacturer stated would be included in the package?: True

What other parts do you consider comparable to this product?: POWER INTEGRATIONS RDK-877

What were the biggest problems encountered?: none

Detailed Review:

Intro and unboxing

RDR-707Q from Power Integrations is a reference design for automotive--graded, non-isolated, 9.75W DC-DC step-down converter, with wide (30V - 550V) voltage input range, fixed 15V output, high efficiency and low power consumption when no load is present.

Design includes automotive grade parts and is capable of operating with -40°C - 85°C ambient air temperature range without heatsink, and even up to 105°C when connected to the enclosure (which is typical to some automotive use-cases, when temperature of enclosure is significantly lower than the ambient air).

Design kit is housed in the elegant cardboard box and contains an PCB with components placed on the one side only - with other side prepared to be screwed to the enclosure.

From the photos one can observe low component count (which decreases overall cost and increases reliability) and PCB design optimized for heat dissipation (optionally through a thermal pad connected to the bottom  side of the PCB).

Documentation

Along with the module, manufacturer provides very detailed documentation, containing not only schematics, but also PCB layout including diagrams of all four layers, detailed BOM, and even assembly tips like inductor placement and a method of manual vias soldering for increased thermal efficiency.

About 40 pages contain performance data in form of measurements, signal shape diagrams on the key components and module outputs during various conditions (both normal operation and fault condition in shape of output short circuit).

Next eleven pages are dedicated to the output voltage ripple measurements - including methodology used and measured signal graphs for various load levels and input voltages.

Last part includes thermal performance (along with thermal camera images) at room temperature, 85°C and 105°C (with thermal pad installed).

Module schematics is a variation of reference design of LinkSwitch-TN2 power supply, but with some modifications (more significant ones marked using red ovals) 

LinkSwitch-TN2 reference design

RDR-707Q schematics

Modifications include:

• additional capacitor (C5) for high frequency filtering,
• R5 resistor for improved output ripple suppression,
• feedback diode constructed as a serial connection of two diodes (D3 and D4) - this modification has dual explanation:
  • some versions of TN2 Design Guide contain the suggestion (according to the PI forum posts) that voltage drop across the feedback diode should be similar to the voltage drop across freewheeling diode drop (but it seems to not be present in the current revision of this document),
  • faster diode improves startup response of the module (RS1M has reverse recovery time of 500ns and UF1JLW about 25ns but is rated at 600V only, forcing serial connection of two diodes),
• R6 and VR1 create conditional pre-load circuit (more about this below).

LinkSwitch-TN2 datasheet states that (describing reference schematics - first one above) "Due to tracking errors between the output voltage and the voltage across C3 at light load or no-load, a small pre-load may be required (R4). For the design in Figure 7, if regulation to zero load is required, then this value should be reduced to 5 kΩ. [...] Recommended values of R4 should be selected to provide a minimum output load of 3 mA. However, lower pre-load resistors increases no-load input power consumption and improves regulation at no-load. Higher pre-load resistors decreases no-load input power consumption and worsen regulation at no-load. Selection of pre-load resistor is a balance between no-load power consumption and no-load regulation."

So - if one expects the load to decrease to (near) zero (or even disconnect), some pre-load is needed (at least to protect power supply components from excessive voltage levels) - some published test results conducted with pre-load disconnected have shown that output voltage can be similar to input level in this case.

But a pre-load consuming 3mA at 15V is wasting 45mW, even under nominal load. Thus, in RDR-707Q another approach was used - assuming that a no-load condition is unlikely and some voltage increase may be allowed in this case, instead of fixed load, R6-VR1 circuit was used. This way, 100Ω pre-load is connected only when output voltage exceeds VR1 threshold (designed to be 20V) and disconnected otherwise.

Power Integrations offer many different resources on their website - one of them being PI Expert, an (online/offline) design helper tool, in which one can - after selecting input parameters, get estimated parameters of key components needed to build a working solution.

As can be seen in RDR-707Q documentation (paragraph  8 - Design Spreadsheet) this design also was initially created using PI Expert, but later optimized. For example, feedback diode was initially designed as single (with Vf of 0.7V), resulting in different RFB (R2) value. This case shows that RDR-707Q was manually optimized for performance.

Other observed deviations from the output of the current revision of PI Expert (with has different version than the one used in the datasheet) are: different main inductor value and possible operation from 30V voltage level (that is causing a warning from the online PI Expert as too low).

I have asked the customer support of the manufacturer how to deal with the observed differences between PI Expert output and the reference design  (very fast response from the support by the way) and they advised for using reference design as actually built, tested and optimized.

Test setup

During this roadtest preparation, two test cases were planned:

• module response for load change,
• module efficiency measurements,

 After analyzing module documentation, I have decided to conduct my tests with different levels of pre-load (most results from the datasheet assume at least 10% of initial load), resulting in the test setup as below:

As can be seen, this circuit allows for dynamic load change (using PWM generator) with different levels of initial load.

All the tests were conducted with 60V input voltage, which allowed for safe evaluation, but probably somewhat limited the possible test scope.

Dynamic parameters

Load change response test was conducted by switching 54Ω load with the frequency of 17Hz and duty cycle of 10%, thus resulting in short load peaks of about 280mA, increasing power supply's load by about 43%.

Pre-load was selected from parallel connection of one, two or three 820Ω resistors, resulting in power supply pre-load of 3% (about 18mA), 5.7% (about 37mA) and 8.5% (about 55mA) - all the values significantly higher than the required 3mA.

3% to 46% load change

Using vertical scaling of 200mV/div, one can observe about 1V (measured as 1060mV) of output voltage change when load changes from 3% to 46% (18mA to 300mA current change).

6% to 50% load change

When pre-load current is increased to about 37mA (5.7% of the rated current of the module), voltage change is reduced to about 740mV when load changes from 6% to about 50% (about 320mA)

8.5% to 52% load change

When pre-load current is increased to 55mA (using 273Ω load), voltage difference further decreases to about 576mV when current changes from 55mA to about 333mA (8.5% to 52% load change)

From the tests conducted, it seems that 3mA pre-load suggestion can be treated as a compromise between additional current draw and output voltage stabilization and the module behaves best when load variability does not include very low values.

Efficiency testing

First, I must admit that efficiency testing didn't go as smooth as I wanted, but I have learned a few things in the process.

Let's start with the basics: module documentation shows the following diagram as a module efficiency measurement setup

Nothing complicated at first glance - but when input current looks like below, things get complicated rather fast

The above graph shows data captured across 27Ω measurement resistor with 820Ω load on the secondary side, 5V/div scale (15.8V peak-to-peak resulting in maximum measured current of about 580mA), with the expected mean current of about 6mA. 

Several issues can be identified:

• meter should have measurement bandwidth wide enough to properly measure a signal with edges that sharp,
• dynamic range is also important to obtain low averaging error when the mean value is about 1/100 of maximum value,
• resistance of the current measurement resistor should be low enough to not introduce voltage drop that would create large measurement error when calculating mean power (for example - in the case above, mean input voltage of the module could be near the nominal voltage of DC input supply [e.g. 60V], but most of the instant power draw would be at voltage about 45V, which could lead to efficiency measurements artificially lowered) 

I have tried different approaches, using several different meters and even employing numeric integration of sample values obtained from the oscilloscope to have instant power values calculated correctly, but the results were lower than efficiency data presented in the module's datasheet. 

Finally I have obtained test setup giving the results similar (but somewhat lower) as in the datasheet. It consisted of two measurement resistors: 2.2Ω (at primary side), 4.68Ω (at secondary) and static load selected from the set ∞Ω (no load), 821Ω, 413Ω, 275Ω, 196Ω and 99Ω. Voltages were measured using MP730889 DMM (which I have won during this year's E14 Experimenting with Extreme Environment Design Challenge). Results are as below:

where:

• Rload is an load resistance connected,
• Vri - voltage across current measurement resistor on primary side,
• Ii - calculated primary side current,
• Vi - voltage at primary side,
• Vro - voltage across current measurement resistor on secondary side,  
• Io - calculated secondary side current,
• Vo - voltage at secondary side,
• Pi - calculated power draw at primary side,
• Po - calculated power draw at secondary side,
• Efficiency - calculated efficiency,
• Load - percentage of maximum load of the module,

Comparing it with efficiency graph from the datasheet (below) we can see the difference of 3-4% in comparable area (initial part of 60V curve doesn't include measurements for loads lower than 10%). This difference can be attributed to the measurement errors - for example, insufficient bandwidth of the meter used or averaging time too short (measurements from the datasheet are using 1 minute averaging window)

Another observation is that under no load output voltage is indeed 20V (limited by R6-VR1 pre-load circuit), and measured no-load input power of 57mW is about two times the value from the datasheet - but it again can be attributed to the measurement error, especially when input current curve looks like below, where all the previously mentioned error sources can have biggest impact on the result:

Summary

In my opinion RDR-707Q is an interesting module, with good parameters and some clever workarounds reducing limitations of this design (active pre-load being engaged only when needed and disconnected otherwise). It has good, detailed documentation, with all sorts of measurements and signal curves, thermal images and construction details. It could be a real help for the designer, beyond the suggestions that can be obtained from PI Expert software.