Characterizing LDO Noise: Part 2

Previously we used the QA350 microvolt DVM to look at the standard deviation of different types of LDOs. That measurement helped us understand the noise performance of the LDOs at 1 Hz and below. This post looks at the same LDOs using the QA401 audio analyzer. The QA401 will let us understand the noise performance of the LDOs from roughly 1 Hz up to 100 kHz. You'll see there is huge difference in a $0.10 LDO versus a $6 LDO and this difference is easily measured with the right equipment. 


The first part isn't an LDO: The LM3670 is a DCDC switching regulator. It's an awesome part for taking 5V down to 3.3V, needing only an inductor and a modest output cap. The LM3670 has two switching modes: PFM and PWM. Pulse Frequency Modulation (PFM) is used by the part to handle loads under 70 mA or so. In this mode, the LM3670 will skip cycles to achieve regulation. As the load demands increase, the effective switching frequency will also increase. Beyond 70 mA, the LM3670 will rely on switching at a constant frequency and modulating the duty cycle to achieve regulation. That mode is the PWM (pulse width modulation) mode of operation.

In the plot below, the QA401 is used to measure the RMS noise of the LM3670 at about 7 mA of load current. From 20 to 96 KHz, we can see the noise is roughly 4.3 mVrms. Notice the spur at 58 KHz: This is the effective switching frequency of this regulator at this very light load (about 2% of the parts rated load). If the load were higher, the peak would shift to the right, ultimately arriving at the fixed frequency (~500 KHz) used for the PWM mode of operation.

This part costs about $1.17. That's higher than an LDO, but this part can deliver north of 300 mA--something the LDOs below cannot. It's a great choice for powering a lot of logic in an FPGA, for example. 


For the MIC5504, the measured noise is about an order of magnitude better than the DCDC we measured above: 316 uVrms. On this plot and subsequent, the noise floor of the QA401 has been overlaid in green so that you can see how much margin the QA401 has to the LDO. In this case, it's nearly 50 dB. 

The cost on this part is almost unreal at $0.11 quantity one. It has a limited input voltage range of 5.5V. This part would be a great choice for powering 3.3V logic from a 5V rail. But you'd not want to use this part as the reference to an ADC or DAC beyond 8 or 10 bits or so.


The LP2985 improves the noise by almost another factor of 10 over the MIC5504 to 40 uVrms. But notice all the "grass" starting at 3 kHz and extending all the way up to 100 kHz. This is substantial and is the result of all the LDOs in this test being supplied by 5V from a USB hub--the same hub powering the QA401. In this case, the LDO is having trouble suppressing the hub noise. 

And while the MIC5504's overall noise was quite a bit worse than the LP2985, you can see the MIC5504's noise rejection was in fact better north of 50 KHz. The primary takeaway here is that the LP2985 offers really good noise performance but suffers in its ability to reject noise at higher frequencies. And this shows up in the spec on the LP2985: A 4.7uF to 10 uF output cap gives you around 35 dB of rejection at 10 kHz. That's pretty poor from a spec standpoint. The LP5907 shown below offers about 70 dB of rejection at 10 kHz and the LT3042 delivers about 90 dB.

The cost on this part is about $0.60 quantity 1. Overall, it's a pretty good part, with extended input range up to about 16V. The extended input range is the primary reason to use this part. It's a great part if you need a few mA from a 12V rail, for example. 



The LP5907 whacks the noise by about a factor of 5 over the LP2985. It's also showing substantial improvement on the USB noise present at the source of the LDOs. The spec claims to offer 70 dB of PSRR at 10 kHz, which is very good. 

The cost on this part is about $0.10 less than the LP2985, but it suffers from a limited voltage range of 5.5V max.If you needed a reference for a 12-bit or higher DAC or ADC, this part would merit consideration. 


Finally, looking at the last of the LDOs, we have the super-premium LT3042, which is about the highest-performing LDO part out there today. In fact, as you can see from the graph, the LT3042 noise is at the limit of the QA401 and to accurately measure this, you'd need a very low-noise 20 dB amp.

This part costs about $5.75 quantity 1. This part is premium in every regard: Killer input votage range (up to 20V), programmable current limiting via single resistor, and a PSRR that continues out to forever. 



The last measurement is of the MAX6126. This isn't an LDO: It's a reference with the ability to source or sink 10 mA or so of current that we also looked at in part 1. The noise isn't quite as good as the LT3042, but it's very respectable still. For accuracy at or below 1 Hz, this part wins everything. And if you were looking to provide a reference for a DC volt meter, this would be a top part to consider (and it's used in the QA350). 

But beyond 100 Hz, the noise performance is fairly similar to the LP5907.

The Test Setup

For all the of tests in part 1 and part 2, the following test setup was used. This test board is powered by USB, and feeds all the regulators in parallel. The DCDC part was disabled by removing the output inductor when testing all the other parts to ensure the switching spur didn't show up in the spectrum. 

On the left side is the USB power entry point, and on the right side are the 5 BNC connectors that allow the regulator outputs to be easily connected to the QA401 or QA350.

There is a single 2.2uF of bulk capacitance at the USB port, and each regulator had 0.1uF at the input and the minimum capacitance specified at the output.

The LP2985 had the noise reduction cap (10 nF) placed.


From Part 1 of this post, the MAX6126 delivered the leading stability in roughly a 1 Hz bandwidth. And that makes sense because it's a reference voltage. But as we look out through the audio band, the engineering achievement of the LT3042 is unmatched.  

Note that it would be fairly difficult or even impossible to understand the performance between the various parts with just a DVM or oscilloscope. Low-noise design requires you to have the right tools that can help you see differences in performance. If your bench instruments can't reveal the difference between a $0.50 LP5907 and a $6 LT3042, think about investing in tools that can highlight these differences for you. 

 If you liked the post you just read, please consider signing up for our mailing list at the bottom of the page. 

Leave a comment

Please note, comments must be approved before they are published