As mentioned in Part I, one concern of the modification described there, is that the output signal has a lot of close-in sidebands. It is suspected that this is the result of using 10 MHz for the reference instead of 25 MHz. To determine whether this is the case or not, this page describes the process of replacing the 10 MHz RFS input reference signal with one closer in frequency to the original 25 MHz - but still derived from the 10 MHz RFS atomic standard signal.
Several methods can be employed to provide higher frequency reference clocks.
A PLL-based frequency multiplier driven by the 10 MHz RFS signal.
A direct frequency-multiplying circuit where the input frequency must be a integral division of the output frequency.
This method is more convenient than the direct frequency multiplication as the input frequency can be one readily available (e.g., 10 MHz references) - requiring only an adjustment to the divider values to arrive at any frequency within the range of the VCO. The aspects to be kept in mind is that there will be a limiting frequency step resolution governed by the number of bits in the internal dividers, and fractional/N dividers (as used in the MAX2870) introduce close-in noise caused by the switching cycle between two integer division ratios which implements the fractional division ratio.
NOTE: it is unclear whether the close-in sidebands observed after replacing the internal 25 MHz crystal oscillator with the external 10 MHz RFS signal is due to the MAX2870 PLL loop values being inappropriate for the 10 MHz input (being 40 % of the 25 MHz design frequency) - or caused by the change in reference frequency from 25 MHz to 10 MHz needing a "noisier" fractional division.
It would also be useful to compare the signal produced when the output frequency set is an exact integer multiple of the input reference frequency, to that when the output frequency is not an integer multiple. For example - using a 10 MHz input reference - set the output to 1528.7 MHz (10 MHz x 152.87 : fractional) and then to 1530.0 MHz (10 Mhz x 153 : integer) and examine the output signal for close-in sidebands in each case.
This method is less convenient than the PLL-based method - but has the advantage of the output phase noise tracking the input phase noise proportional to the multiplication ratio (phase noise ratio (dB) = 20 log(n) where 'n' is the multiplication ratio). This advantage can turn into a disadvantage w.r.t. PLL designs at higher multiplication ratios. A typical design for direct frequency multiplication is to use the input signal to generate a square wave (i.e., generating harmonics) and then use series crystals to pass the desired harmonic - which will be thusly restored to sinusoidal. A number of design solutions for this method present themselves.
Using the 10 MHz RFS sinewave signal to output 30 MHz (the 3-rd harmonic) - which, while not the desired 25 MHz, is nonetheless closer than 10 MHz is.
Using the 10 MHz RFS sinewave signal, divide by 2 (giving 5 MHz) and then extract the 5-th harmonic to give the desired 25 MHz.
Re-program the RFS to output (25 / 3) MHz - i.e., 8.333333..... MHz to output 25 MHz energy (3-rd harmonic). There is an RFS on hand which can be re-programmed to that frequency. It is known that this output frequency is within the working range of the RFS as they can come programmed as 1 PPS generators - which requires setting the RFS to 8.388608 MHz to feed to an internal 23-bit divider (8.388608 MHz / 2^23 = 1 Hz).
Which approach is finally adopted will depend on results of the following investigations.
To gather information about the behaviour of the signal generator w.r.t. presence of sidebands when driven by different sources the following arrangements are tested and compared. The signal monitored is the output of the signal generator when configured to receive the relevant input reference frequency (10 MHz, 30 MHz, and 25 MHz) and set to 1528.7 MHz (which produces harmonics up to and including the required 12.2 GHz Ku-Band test signal).
Un-modified signal generator - onboard 25 MHz crystal oscillator.
The first modification - feeding a 10 MHz RFS signal (as used in Part I) as the reference signal.
Using the 10 MHz RFS sinewave signal to output 30 MHz (the 3-rd harmonic) by direct frequency multiplication to use as the reference signal.
As the signal generator had already been modified by removing a 0 ohm resistor (Modifying Signal Generator to External Reference - Part I ) and previous pre-modification results were gathered using a slightly different test setup, the signal generator was temporarily 'un-modified' by a solder bridge to re-connect the onboard 25 MHz crystal oscillator. The result for an output frtequency of 1528.700 MHz with the original onboard crystal oscillator is shown on the right. Apart from the measurement artifact towards the right, the output is relatively clean.
The onboard 25 MHz reference oscillator was disconnected by removing a 0 ohm resistor (R7) which was serving as a link. The original MCLK output of the onboard crystal oscillator is now used as an external reference input.
Various external reference signal sources were tested.
The results from when a 10 MHz RFS is connected directly to the MCLK input is shown on the right. It was thought that the sideband spurii is caused by insufficient drive level of the RFS output, but a check of levels and waveforms showed the signal level and waveform are roughly equivalent. To cross-check, signals from a Fluke 6060AN Signal Generator were applied.
10 MHz Rubidium Frequency Source Reference
Various reference frequencies and drive levels were tested using a Fluke 6060AN signal generator.
Although the 'fur' sidebands seen with the RFS were not present, the spread of signal energy was still evident in the two examples (10 MHz/+10 dBm and 20 MHz/+10 dBm) as shown on the right. The minimum level for the external reference input was found to -17 dBm. This shows that the approx. +6 dBm amplitude from the RFS is well above that required and so the 'fur' seen in the RFS result is not due to insufficient drive.
All drive level variations tested with the 6060AN showed this same pattern of spread.
For reasons unknown the MAX2870 reference input does not like being driven from either the 10 MHz RFS source or the Fluke 6060AN. The fact that the onboard 25 MHz crystal oscillator results in a clean 1.5 GHz signal may mean that the MAX2870 may be sensitive to the nature of noise on those two sources - although no evidence for such noise could be found in those sources.
Out of curiosity, a second MAX2870 signal generator with its original 25 MHz crystal oscillator (i.e., unmodified) was used as the first (modified) MAX2870 signal generator's external reference. Pleasingly this showed a nice clean 1.5 GHz output signal.
Unfortunately, the accuracy and stability of this arrangement falls back to the level of the onboard 25 MHz crystal oscillator. But this result triggered an alternative idea.
From the good results obtained from using a second MAX2870, the idea arose to use one of those inexpensive 'frequency multiplier' boards (AliExpress USD$12) to take the RFS 10 MHz signal as input and produce a 25 MHz output. Unfortunately the ones on hand (using the NB3N501 IC) only have jumper settings for - among other ratios - a factor of x2, x3 and x5. The second one of these two units will be modified to using an NB3N502, which has among its selections a factor of x2.5 - perfect for using a 10 MHz RFS to derive a 25 MHz reference signal.
However, in the meantime before that modification is done, the unmodified unit was tested - giving 20 MHz, 30 MHz and 50 MHz references driven by the 10 MHz RFS signal. The results of those three reference frequencies is shown on the right (20 MHz) and below (30 MHz and 50 MHz) - left to right.
NOTE: this method is NOT direct frequency multiplication as the board has a number of non-integer ratios (e.g., 3.125) - so it appears to be a fractional/N PLL loop design. This may be fortuitous as direct multiplication might carry through the particular characteristics of the RFS and Fluke signals which seem to cause 'fur'. By PLL-type multiplication the output signal comes indirectly from an onboard VCO - which could 'clean up' the incoming RFS signal.
10 MHz RFS Multiplied to 30 MHz
10 MHz RFS Multiplied to 50 MHz
By using the Rubidium Frequency Source (RFS) as the primary reference and multiplying its 10 MHz output to - say - 30 MHz, and then using that clock as the external reference for a MAX2870 Signal Generator, an output signal in the range from 23 MHz to 6 GHz is available with the stability of the RFS.
As the motivation to get a stable frequency signal for the MAX2870 signal generator is to test the drift of Ku-Band LNBF local oscillators, the stability requirements for this application have been met.
The frequency accuracy of these signals is determined by the accuracy of the Fractional/N dividers of the MAX2870 and is tested in MAX2870 Sig Gen Frequency Accuracy .
The above signal quality might be of concern if used for actual observing or measuring, say, system temperature - where the noise sidebands might negatively affect the S/N depending on the particular use. However, in this case the concern about sidebands arose simply because, in the cases shown above of direct RFS or 6060AN injection, the resultant spread in the signal makes for unreliable measurement of the frequency and frequency drift. The use of the clock multiplier eliminates that gross unreliability and returns the level of noise sidebands to that which occurs when using the original onboard 25 MHz crystal oscillator.