Topics

si5351 crosstalk #radiuno

ajparent1/KB1GMX
 

Another stone turned.

While testing VFO/BFO replacement for an Atlas210 I came across a spur at the BFO frequency 
while looking at the VFO signal quality.  I first figured it was layout or some error.  Sadly it was not.
As the Atlas210 was a high performance radio for ts day this would compromise it.  For that the
crystal oscillator BFO goes back in.

So I went to the Raduino and did some research and built two other commercial SI5351 breakout
boards to test for CROSSTALK.   Indeed if you load (resistive load) a oscillator port you will see
it as a weak spur on the other operating port.  The amount of the spur is proportional to the
load current for that oscillator.  Its chip level not board level.

Two oscillators 0 and 2 operating with 0 set for 4.0000mhz and 2 set for 5.520mhz.

The output hooked to SA via 3db pad.  Its output is measured as 8.94 DBM, I expected 13!
Not a big problem and clean.  Add a 50 ohm laod to osc-2 and there it is at 5.520mhz, a
signal exactly 58dbc.  Move SA to osc-2 and big as life and 8.94 DBM again (where is the 13dbm?)
and sure enough a -58dbc spur at 4.0000mhz.  Take the load off the 4mhz osc-0 and
the spur drops into the noise about -90dbc down.

Seems running multiple oscillator lowers the output power of any one too.

Make the load resistor 25ohms and it gets about 6db worse and at 240ohms its about -70dbc.

These do not effectively cause spurs of consequence for TX output. 

However this does help explain why an external crystal osc was needed to get
carrier leakage to a expected level.  Also where the receiver birdies are coming
from.  Its not the Raduino its the 5351 chip.   This makes it a poor LO replacement
unless its the sole operating oscillator as even the lowly AD9850 DDS 
has about the same SFDR (spurious free dynamic range).  IF only one oscillator is
in use the signal has no spurs only harmonics.

In short for higher performance radios only use one oscillator in the 5351 and
use multiple 5351s is you need more oscillators.

Allison

Gordon Gibby
 

What a cool discovery!


On Sep 5, 2018, at 15:31, ajparent1/KB1GMX <kb1gmx@...> wrote:

Another stone turned.

While testing VFO/BFO replacement for an Atlas210 I came across a spur at the BFO frequency 
while looking at the VFO signal quality.  I first figured it was layout or some error.  Sadly it was not.
As the Atlas210 was a high performance radio for ts day this would compromise it.  For that the
crystal oscillator BFO goes back in.

So I went to the Raduino and did some research and built two other commercial SI5351 breakout
boards to test for CROSSTALK.   Indeed if you load (resistive load) a oscillator port you will see
it as a weak spur on the other operating port.  The amount of the spur is proportional to the
load current for that oscillator.  Its chip level not board level.

Two oscillators 0 and 2 operating with 0 set for 4.0000mhz and 2 set for 5.520mhz.

The output hooked to SA via 3db pad.  Its output is measured as 8.94 DBM, I expected 13!
Not a big problem and clean.  Add a 50 ohm laod to osc-2 and there it is at 5.520mhz, a
signal exactly 58dbc.  Move SA to osc-2 and big as life and 8.94 DBM again (where is the 13dbm?)
and sure enough a -58dbc spur at 4.0000mhz.  Take the load off the 4mhz osc-0 and
the spur drops into the noise about -90dbc down.

Seems running multiple oscillator lowers the output power of any one too.

Make the load resistor 25ohms and it gets about 6db worse and at 240ohms its about -70dbc.

These do not effectively cause spurs of consequence for TX output. 

However this does help explain why an external crystal osc was needed to get
carrier leakage to a expected level.  Also where the receiver birdies are coming
from.  Its not the Raduino its the 5351 chip.   This makes it a poor LO replacement
unless its the sole operating oscillator as even the lowly AD9850 DDS 
has about the same SFDR (spurious free dynamic range).  IF only one oscillator is
in use the signal has no spurs only harmonics.

In short for higher performance radios only use one oscillator in the 5351 and
use multiple 5351s is you need more oscillators.

Allison

Jerry Gaffke
 

Not unexpected.
Those are primarily CMOS logic outputs, operating from a single Vcc pin.
And not much good at driving into 50 ohms:
    https://groups.io/g/BITX20/search?q=si5351+crosstalk+ke7er

Maybe need a stuffing option on the Raduino to have 3 different si5351's?
Though possible we can get by with one if we load it lightly, adding an emitter-follower to each output.

Jerry


On Wed, Sep 5, 2018 at 12:31 PM, ajparent1/KB1GMX wrote:
Another stone turned.

While testing VFO/BFO replacement for an Atlas210 I came across a spur at the BFO frequency 
while looking at the VFO signal quality.  I first figured it was layout or some error.  Sadly it was not.
As the Atlas210 was a high performance radio for ts day this would compromise it.  For that the
crystal oscillator BFO goes back in.

So I went to the Raduino and did some research and built two other commercial SI5351 breakout
boards to test for CROSSTALK.   Indeed if you load (resistive load) a oscillator port you will see
it as a weak spur on the other operating port.  The amount of the spur is proportional to the
load current for that oscillator.  Its chip level not board level.

Two oscillators 0 and 2 operating with 0 set for 4.0000mhz and 2 set for 5.520mhz.

The output hooked to SA via 3db pad.  Its output is measured as 8.94 DBM, I expected 13!
Not a big problem and clean.  Add a 50 ohm laod to osc-2 and there it is at 5.520mhz, a
signal exactly 58dbc.  Move SA to osc-2 and big as life and 8.94 DBM again (where is the 13dbm?)
and sure enough a -58dbc spur at 4.0000mhz.  Take the load off the 4mhz osc-0 and
the spur drops into the noise about -90dbc down.

Seems running multiple oscillator lowers the output power of any one too.

Make the load resistor 25ohms and it gets about 6db worse and at 240ohms its about -70dbc.

These do not effectively cause spurs of consequence for TX output. 

However this does help explain why an external crystal osc was needed to get
carrier leakage to a expected level.  Also where the receiver birdies are coming
from.  Its not the Raduino its the 5351 chip.   This makes it a poor LO replacement
unless its the sole operating oscillator as even the lowly AD9850 DDS 
has about the same SFDR (spurious free dynamic range).  IF only one oscillator is
in use the signal has no spurs only harmonics.

In short for higher performance radios only use one oscillator in the 5351 and
use multiple 5351s is you need more oscillators.

Allison

 

 

Gordon Gibby
 

Perhaps some connection through chip geometry or through power supply or ground systems. 


On Sep 5, 2018, at 15:35, Gordon Gibby <ggibby@...> wrote:

What a cool discovery!


On Sep 5, 2018, at 15:31, ajparent1/KB1GMX <kb1gmx@...> wrote:

Another stone turned.

While testing VFO/BFO replacement for an Atlas210 I came across a spur at the BFO frequency 
while looking at the VFO signal quality.  I first figured it was layout or some error.  Sadly it was not.
As the Atlas210 was a high performance radio for ts day this would compromise it.  For that the
crystal oscillator BFO goes back in.

So I went to the Raduino and did some research and built two other commercial SI5351 breakout
boards to test for CROSSTALK.   Indeed if you load (resistive load) a oscillator port you will see
it as a weak spur on the other operating port.  The amount of the spur is proportional to the
load current for that oscillator.  Its chip level not board level.

Two oscillators 0 and 2 operating with 0 set for 4.0000mhz and 2 set for 5.520mhz.

The output hooked to SA via 3db pad.  Its output is measured as 8.94 DBM, I expected 13!
Not a big problem and clean.  Add a 50 ohm laod to osc-2 and there it is at 5.520mhz, a
signal exactly 58dbc.  Move SA to osc-2 and big as life and 8.94 DBM again (where is the 13dbm?)
and sure enough a -58dbc spur at 4.0000mhz.  Take the load off the 4mhz osc-0 and
the spur drops into the noise about -90dbc down.

Seems running multiple oscillator lowers the output power of any one too.

Make the load resistor 25ohms and it gets about 6db worse and at 240ohms its about -70dbc.

These do not effectively cause spurs of consequence for TX output. 

However this does help explain why an external crystal osc was needed to get
carrier leakage to a expected level.  Also where the receiver birdies are coming
from.  Its not the Raduino its the 5351 chip.   This makes it a poor LO replacement
unless its the sole operating oscillator as even the lowly AD9850 DDS 
has about the same SFDR (spurious free dynamic range).  IF only one oscillator is
in use the signal has no spurs only harmonics.

In short for higher performance radios only use one oscillator in the 5351 and
use multiple 5351s is you need more oscillators.

Allison

ajparent1/KB1GMX
 

Jerry,

Yep been around cmos design and likely a bus common as well as substrate current paths.
I have to try an emitter follower or fet based amp to unload it as its not a good thing.

Allison

ajparent1/KB1GMX
 

On Wed, Sep 5, 2018 at 12:35 PM, Gordon Gibby wrote:
What a cool discovery!

I knew it was there but never saw it quantified or even measured while digging on the net.
That and no one seemed to know it was loading sensitive.

I suspect a CMOS buffer (there are many and cheap) would be a good solution.
Three of them would cost near nothing and do a good job.

Allison

Gary Anderson
 

When the uBiTX spurs were brought to public attention, I dug up this nice little write up on the si5351.
https://nt7s.com/tag/crosstalk/

Also note the Si data sheet for decoupling recommendations and unused pins in section 6.

Digital design and layout on the Raduino could be improved here too if revisited, although not the biggest of issues at hand.
(e.g. I2C pull-ups on the wrong side of the decouple caps per the schematic)

Regards,
Gary

Glenn
 

It's a shame the manufacturers didn't put 2 more pins on the chip and use them as I2C address select lines.

Right now, the only way i can see to use multiple (10pin) Si5351's in a rig is to have say, an ATtiny or similar 8 pin uP to program it on power up. Or as commanded by the master uP.

glenn

Gary Anderson
 

I can see why the manufacture doesn't  Added cost for a very unique implementation.
I2C mux is a solution if you need multiple same address devices on the same I2C bus. 

What I haven't researched is if the si5351 with higher clock count (8 output variants) with only 2 clocks per VDDO have lower cross talk using 1 clock per VDDO.
Since data shown is load dependent, it suggests that it would be.

Regards,
Gary

MadRadioModder
 

Use this I2C MUX.  I’ve used it before with great success…

 

https://www.nxp.com/docs/en/data-sheet/PCA9547.pdf

 

MRM

 

From: BITX20@groups.io [mailto:BITX20@groups.io] On Behalf Of Glenn
Sent: Wednesday, September 5, 2018 5:11 PM
To: BITX20@groups.io
Subject: Re: [BITX20] si5351 crosstalk #radiuno

 

It's a shame the manufacturers didn't put 2 more pins on the chip and use them as I2C address select lines.

Right now, the only way i can see to use multiple (10pin) Si5351's in a rig is to have say, an ATtiny or similar 8 pin uP to program it on power up. Or as commanded by the master uP.

glenn


Virus-free. www.avg.com

--

…_. _._

ajparent1/KB1GMX
 

There are a raft of bidirectional switches that would work for two or 3.

Allison

Ashhar Farhan
 

All may not be lost. The spurs start as you start to draw power. So, don't draw power. A source/emitter follower will lightly load the si5351 outputs. Buffering helps.
There are other ways to use multiple si5351s. A simply solution is to use analog muliplexers to switch the i2c lines between multiple devices. 
- f
- f


On Thu, 6 Sep 2018, 05:36 ajparent1/KB1GMX, <kb1gmx@...> wrote:
There are a raft of bidirectional switches that would work for two or 3.

Allison

ajparent1/KB1GMX
 

Farhan,

Testing says buffers work The SN74AUP1T04 inverter one per package
(4 pins like the mmbt3904s) or the 74LVT04 (6 per)work well.

I can do that with transistor or Fet but then I need bias and other parts
as both have thresholds to hit and we want them to work at 80mhz
or higher.

That alone makes it harder to use just cheap devices.

Allison

rcbuck@...
 

Allison,

I used a Si5351A in a 20/30/40 meter receiver I recently built. The crystal filter frequency was 9 MHz. So my LO frequencies were 16.0-16.3 MHz, 19.0-19.15 MHz, and 23.0-23.35 MHz. I used low pass filters on the LO output to knock out the harmonics. The 20 and 30 meter bands share a common LPF with a cutoff around 19.2 MHz. The BFO output has a 9 MHz bandpass filter to clean up the output. I used output 0 for the LO and output 2 for the BFO. I never thought to look for the 9 MHz signal in the LO output. Thanks for discovering this problem and the simple solution for it.

Sometime in the next couple of weeks I will open the receiver up and put the LO output on the SA to see if the 9 MHz signal is present. The LO (channel 0) is operated at 2 mA output and drives a feedback amplifier. There is a 3 dB, 50 ohm pad on the output of the amp which results in just over 7 dBm into the ADE-1 first mixer. The BFO output is set to 6 mA output and drives an ADE-1 product detector directly. My guess is the 9 MHz shows up in the LO output but the LO will not show up in the BFO output due to the band pass filter.

The cost to build this receiver was more than the uBitx costs. You can tell that by looking at the attached pdf. All filters, crystal, band pass, and low pass used separate relays for the input and output sides. Lessons learned a long time ago.

Ray, AB7HE

VK3HN
 

Ray, that has to be the best shielded homebrew rx I think I've ever seen!  And SMAs everywhere. I bet there isn't a sniff of RF anywhere within a mile of your receiver.   Congratulations!

Lawrence Galea
 

Congratulations to Allison and the rest for all their trouble shooting and discoveries.
Regards

On Thu, Sep 6, 2018 at 8:04 AM VK3HN <prt459@...> wrote:
Ray, that has to be the best shielded homebrew rx I think I've ever seen!  And SMAs everywhere. I bet there isn't a sniff of RF anywhere within a mile of your receiver.   Congratulations!

Hans Summers
 

Hi Glenn

> Right now, the only way i can see to use multiple 
> (10pin) Si5351's in a rig is to have say, an ATtiny or
> similar 8 pin uP to program it on power up. Or as
> commanded by the master uP.

It is not difficult to program a second (or third etc) firmware-emulated I2C port on the microcontroller. The inbuilt I2C/TWI peripheral isn't the only way to do it. I have code examples in the pages below http://qrp-labs.com/synth

This way, you can program multiple Si5351A from the same microcontroller. You need only one additional I/O pin per device since all Si5351A can share the same SDA pin. Only SCL needs to be unique per device. 

However in my opinion this is a somewhat complicated solution to the problem. 

In many situations there is NO problem. You may not be using multiple Si5351A outputs concurrently. The spurs may not be on frequencies that matter, in the design. Or the spurs may be at a neglible level.

The last point is important. The spurs are worse the lower the impedance of the load on the Si5351A outputs. I think that this is entirely expected within chip, but it is very nice to see it quantified.

The easy solution is don't load the outputs heavily. Use a buffer with a high impedance input. Then the spurs disappear into the noise floor. In most cases this solution is probably cheaper and easier than multiple Si5351A.

73 Hans G0UPL 

ajparent1/KB1GMX
 

On Wed, Sep 5, 2018 at 07:45 PM, <rcbuck@...> wrote:
I used a Si5351A in a 20/30/40 meter receiver I recently built. The crystal filter frequency was 9 MHz. So my LO frequencies were 16.0-16.3 MHz, 19.0-19.15 MHz, and 23.0-23.35 MHz. I used low pass filters on the LO output to knock out the harmonics. The 20 and 30 meter bands share a common LPF with a cutoff around 19.2 MHz. The BFO output has a 9 MHz bandpass filter to clean up the output. I used output 0 for the LO and output 2 for the BFO. I never thought to look for the 9 MHz signal in the LO output. Thanks for discovering this problem and the simple solution for it.

Ray,

Shielding costs weight and money is no surprise, it also gets good results!  Using the 5351 as a single
oscillator is actually not bad it when you try to use it for but LO and CIO the fun begins.  Solution use
a buffer that presents a high impedance to the 5351.  Lower the circulating current at the outputs and
the undesired cross-talk goes down.

Band pass filter can limit the impact of this but buffering first really helps.  I tried a 74LVT04 and the
spur was near gone (down in the HP8568B SA noise).  Simple fix as those are available as one
inverter per sot package.  They are cheap too Mouser gets 37 cents for them for qant=1.
So buffer then band pass you get even better result.

Me I cut my teeth doing superhet radios starting in the late 60s and the first thing to overcome was
unstable VFOs.  By time I started college I'd developed a deep understanding of PLL systems and
had made one for a radio using TTL/CMOS and whatever was available in 1971.  So many issues
of synthesized RF sources are seriously old hat.

What shocked me is after doing research and experiments I looked at many designs and noted
very few people understood this or paid attention to it.  Interesting thing is the only ones that 
do not have issues either use the Sa612 (high input impedance) and the QSD mixers (cmos switch).

Allison

Gordon Gibby
 

​wow, the free knowledge just keeps flowing!

cheers, 

gordon



From: BITX20@groups.io <BITX20@groups.io> on behalf of ajparent1/KB1GMX <kb1gmx@...>
Sent: Thursday, September 6, 2018 12:33 PM
To: BITX20@groups.io
Subject: Re: [BITX20] si5351 crosstalk #radiuno
 
On Wed, Sep 5, 2018 at 07:45 PM, <rcbuck@...> wrote:
I used a Si5351A in a 20/30/40 meter receiver I recently built. The crystal filter frequency was 9 MHz. So my LO frequencies were 16.0-16.3 MHz, 19.0-19.15 MHz, and 23.0-23.35 MHz. I used low pass filters on the LO output to knock out the harmonics. The 20 and 30 meter bands share a common LPF with a cutoff around 19.2 MHz. The BFO output has a 9 MHz bandpass filter to clean up the output. I used output 0 for the LO and output 2 for the BFO. I never thought to look for the 9 MHz signal in the LO output. Thanks for discovering this problem and the simple solution for it.

Ray,

Shielding costs weight and money is no surprise, it also gets good results!  Using the 5351 as a single
oscillator is actually not bad it when you try to use it for but LO and CIO the fun begins.  Solution use
a buffer that presents a high impedance to the 5351.  Lower the circulating current at the outputs and
the undesired cross-talk goes down.

Band pass filter can limit the impact of this but buffering first really helps.  I tried a 74LVT04 and the
spur was near gone (down in the HP8568B SA noise).  Simple fix as those are available as one
inverter per sot package.  They are cheap too Mouser gets 37 cents for them for qant=1.
So buffer then band pass you get even better result.

Me I cut my teeth doing superhet radios starting in the late 60s and the first thing to overcome was
unstable VFOs.  By time I started college I'd developed a deep understanding of PLL systems and
had made one for a radio using TTL/CMOS and whatever was available in 1971.  So many issues
of synthesized RF sources are seriously old hat.

What shocked me is after doing research and experiments I looked at many designs and noted
very few people understood this or paid attention to it.  Interesting thing is the only ones that 
do not have issues either use the Sa612 (high input impedance) and the QSD mixers (cmos switch).

Allison

Ashhar Farhan
 

For higher end designs, it is best that there as few clocks used as possible. I prefer to use individual Si570s For the tracking generator of my specan, i used two si570 with an relay to switch the i2c lines.
Even then, the isolation was not good enough until used rf shielding on the power lines and the i2c was decoupled extensively. It was quite an effort, at the end, i was still unsatisfied and never published the tracking generator. 
- f

On Thu, 6 Sep 2018, 22:18 Gordon Gibby, <ggibby@...> wrote:

​wow, the free knowledge just keeps flowing!

cheers, 

gordon



From: BITX20@groups.io <BITX20@groups.io> on behalf of ajparent1/KB1GMX <kb1gmx@...>
Sent: Thursday, September 6, 2018 12:33 PM
To: BITX20@groups.io
Subject: Re: [BITX20] si5351 crosstalk #radiuno
 
On Wed, Sep 5, 2018 at 07:45 PM, <rcbuck@...> wrote:
I used a Si5351A in a 20/30/40 meter receiver I recently built. The crystal filter frequency was 9 MHz. So my LO frequencies were 16.0-16.3 MHz, 19.0-19.15 MHz, and 23.0-23.35 MHz. I used low pass filters on the LO output to knock out the harmonics. The 20 and 30 meter bands share a common LPF with a cutoff around 19.2 MHz. The BFO output has a 9 MHz bandpass filter to clean up the output. I used output 0 for the LO and output 2 for the BFO. I never thought to look for the 9 MHz signal in the LO output. Thanks for discovering this problem and the simple solution for it.

Ray,

Shielding costs weight and money is no surprise, it also gets good results!  Using the 5351 as a single
oscillator is actually not bad it when you try to use it for but LO and CIO the fun begins.  Solution use
a buffer that presents a high impedance to the 5351.  Lower the circulating current at the outputs and
the undesired cross-talk goes down.

Band pass filter can limit the impact of this but buffering first really helps.  I tried a 74LVT04 and the
spur was near gone (down in the HP8568B SA noise).  Simple fix as those are available as one
inverter per sot package.  They are cheap too Mouser gets 37 cents for them for qant=1.
So buffer then band pass you get even better result.

Me I cut my teeth doing superhet radios starting in the late 60s and the first thing to overcome was
unstable VFOs.  By time I started college I'd developed a deep understanding of PLL systems and
had made one for a radio using TTL/CMOS and whatever was available in 1971.  So many issues
of synthesized RF sources are seriously old hat.

What shocked me is after doing research and experiments I looked at many designs and noted
very few people understood this or paid attention to it.  Interesting thing is the only ones that 
do not have issues either use the Sa612 (high input impedance) and the QSD mixers (cmos switch).

Allison