-120dBm at 1.5MHz. 100kHz noise improves by
10dB if the output capacitance is raised to 100µF,
but LF noise needs even more to reduce it.
was needed to reduce it by a further 10dB in the
Input noise was measured as -100dBm at
100kHz and could be reduced by increasing the
input capacitor to 100µF, as expected. BUT, this
noise on the
terminal was made worse by
any increase in the
So both capacitors have to be varied together.
shows the result for 100µF on the output
and 10µF on the input. A further 2200µF had to
be placed on the input to reduce noise to close to
the receiver’s measurement limit.
The device oscillated in the LF region with
anything less than 1µF on the output and 470nF
on the input.
The datasheet specification says
“1µF on the input if spaced far
from the PSU filter; 0.1µF on
the output is optional to improve
transient response with typically
1µF to 1000µF as needed”.
0.68µF was used here.
Noise is worst at low
frequencies, with a definite peak
at 84kHz as shown in
But the noise does roll off faster
with this device, dropping to
-110dBm at 300kHz, -120dBm
at 500kHz and down to the
measurement limit at 1.2MHz.
Adding extra output C shifts the
LF peak lower: a value of 100µF
drops the noise at 100kHz to
-110dBm, while 2200µF takes
it nearly to the measurement
Although not normally
advised, some applications
bypass the adjustment pin to
ground to reduce noise. Testing
this with an extra 0.68µF in
that position flattened the low
frequency response of Figure 3
at a level of -100dBm over
a wide band. Proper noise
reduction and decoupling in
the LM317 can become quite
complex, but one thing that
be done is just to
connect a capacitor between
output and the adjustment pin.
A capacitor here will
make it go unstable
from this scenario, the LM317
could otherwise never be made
to go unstable and oscillate.
All three voltage regulators
appreciable levels of noise at
LF to MF frequencies when using the minimum
decoupling specified in the manufacturer’s
datasheets. It is easy to see why they give their
greatest problems with sensitive PLL circuitry and
are also regarded badly in the Hi-Fi audio world.
Of the two fixed voltage ones, the low dropout
LM1117 fared quite a bit worse at LF but did roll
off slightly faster as frequency rose into the HF
region. The adjustable LM317 with minimum
decoupling fared badly at lower frequencies, but
rolled off faster above 300kHz. It also cleaned
up the best of the lot with extra C added on the
output – exactly as the datasheet suggested would
To properly clean up the resulting supplies
requires large values of additional capacitors on
the output. Some extra series resistance and/or a
choke and a large C to form a low pass filter are
going to be needed in critical applications.
Noise appearing on the regulator input pin
is seen in some devices but should not cause
a problem unless the input supply line goes
elsewhere to sensitive circuitry, where conducted
noise can leak through. This was the issue seen
during the loop amplifier tests where the DC input
to the LDO was provided ‘up the coax’. The noise
got into the RF feed to the receiver. A bigger input C
and optionally a series inductor improved matters.
Paul, M1CNK, who was one of the
experimenters who found the noise on the loop
amplifier, commented that “…the main issue,
apparently, is with noise coming from the voltage
reference. Therefore the better [low noise]
regulators have a separate pin for decoupling the
voltage reference. That way the capacitance used
for decoupling is not inside the control loop. One
problem with just adding larger C to the input and
output is that these become part of the control
loop. Hence by adding too much, you slow down
the response of the regulator to noise spikes – so
they just pass through”.
 All noise levels were measured in dBm; the power
delivered into the 50Ω input of the SDR-IQ in a
measurement (FFT) bandwidth of 24Hz. The SDR-IQ
has a minimum sensitivity of around -128dBm at this
setting but a lot of local computer interference is picked
up at frequencies below 100kHz so this sensitivity could
only be achieved above 500kHz. Fortunately (or not)
the voltage regulators under test exceeded even this
pickup level when using the minimum recommended
decoupling. The QRM was also spiky, so ‘real’ noise could
be discerned between the spikes.
To convert a measured dBm reading to a voltage, first
normalise from the 24Hz measurement bandwidth by
subtracting 10.LOG(24) = 14dB from the dBm value.
Then calculate the voltage needed across a 50Ω resistor
to give this power. -100dBm measured = -114dBm/Hz
watts per Hz. In 50Ω this corresponds to
450nV per root-Hz.
 The tantalum capacitor was hand-held in my fingers
while it was placed across the regulator terminal pins for
testing. I got it the wrong way round. It is at times like
that you really appreciate what happens when a tantalum
bead capacitor is connected with the wrong polarity!
 It is easy to fall into this trap unintentionally, as a
capacitor directly across these two pins is often seen with
7805 type devices. So it can be tempting to just place
one when designing and laying out PCBs. The story may
be apocryphal, but a certain spacecraft manufacturer on
the south coast apparently fell into this trap many years
ago. It was only when a PSU badly failed EMC testing
(fortunately while still at the breadboard stage) that the
mistake was found.
Andy Talbot, G4JNTandy.firstname.lastname@example.org
Output noise from the LM1117 LDO regulator when
decoupled with 100µF on the output and 10µF on the input. Input
and output decoupling capacitors have to be increased together to
control noise from this device.
LM317 adjustable regulator with the minimum
specified values of 1µF on the input and 680nF on the output.
The noise peak at 84kHz reduces in frequency as the output
decoupling C is increased, but values up to 1000µF or more are
needed to get it right down to insignificant levels.