Connor-Winfield HTV-series TCXO

LeapSecond.com :: Museum of Time & Frequency

Topics: TCXO performance, GPS disciplined clock, TCXO tempco, Connor-Winfield, ADEV, MTIE.

Introduction

Although I usually work with single or double-oven quartz oscillators (OCXO) and atomic frequency standards I had a chance to see how well a modern TCXO (Temperature Compensated Quartz Oscillator) performs. These are tiny, 14-pin DIP sized, hermetically sealed oscillators made by Connor-Winfield as part of their TCXO product line. The P/N is HTV2.

Typical of a commodity crystal, the HTV2 is specified only with an overall accuracy and drift specification; in this case 1 ppm from 0 to 50 C and 1 ppm / year drift. No phase noise or stability information (e.g., Allan Deviation) is given by the manufacturer. So we'll test 'em and see how well they do. 1 ppm is not bad for a little can oscillator but what if the oscillator is used in a stable environment rather than across the full range of 0 to 50 C? What if the oscillator is to be used for a GPS disciplining application and long-term drift is not an issue?

Test Setup

I received a PCB with four mounted HTV2 oscillators and powered it with a stable 5.000 VDC source (Agilent 6612C). The PCB current consumption at 5 V was 59.5 mA, about 15 mA per oscillator (75 mW).

The PCB buffers the TCXO output with a pair of 'HC04 inverters and a pseudo +/- 5V differential output is taken from across the second inverter. For ground isolation I passed this output through a 50 ohm Mini-Circuits - FTB-1 - RF Transformer. The unbalanced output of the transformer goes to my frequency measurement system.

Actually, the output buffering is done via:
                         |\
                    /----| >o------ Out-
Osc ---*-jumper-*--<  |\ |/  |\
                    \-| >o---| >o-- Out+
                      |/     |/

I had hoped that loading the first inverter less
would reduce the skew, but there's still some.

These oscillators are actually VCTCXO not plain TCXO but for this initial test we'll leave the EFC input alone.

Note these are 10.48576 MHz oscillators.

I placed the PCB inside a 10³ cm anti-static foam insulated box to reduce thermal variations due to airflow near my desk in the lab. The box is not temperature controlled or regulated; it simply provides some physical isolation and thermal mass. Later on we can systematically check the performance over temperature but right now we're more interested in a quick look at short- and mid-term stability.

Below are some photos of the setup.

For scale, an HP 10811 OCXO inside my homemade thermal isolation box

The TCXO, thermal isolation box, and 5 VDC power supply (top)

Four TCXO mounted on PCB inside thermal isolation box

Ground isolation of differential output

Checking the output waveform under 50R load

Here are 'scope traces of the output waveform, without and with 50R termination. Although a 'scope or frequency counter allows 1M input impedance, my time analyzer is 50R only.

PCB -> RF transformer -> scope (1M)

PCB -> RF transformer -> 50R -> scope (1M)

Close-up of ~5 ns rise-time

Power supply, TCXO box, oscilloscope, messy lab, etc.

Time Interval analyzer, frequency standards, GPS, messy lab, etc.

First Results

The following plots were made a little over half a day into a clean measurement run.

Conner-Winfield HTV2 #1 TCXO, 50k seconds of data

Plot notes (02-Mar-2003 Sunday evening / Monday morning run):

These plots were taken about 14 hours into the run (50,000 seconds).
TSC 5110A phase/frequency plots show the most recent 10 minutes only.
Channel A 5 MHz reference is active H-Maser.
The frequency at run start was 5 MHz * B/A = 10.485 778 723 MHz.
The net time drift after 50k seconds is about -150 µs.
The net frequency drift after 50k seconds is about +8.8E-9.
The EFC voltage was constant but not recorded.
The temperature was probably constant to a few degrees F but not recorded.
The raw data for this run is log12211.txt (50071 lines, 681 KB). This is time interval data (phase difference) in 200 ns units.

Below are phase and frequency plots using the entire data set and also ADEV, MDEV, and MTIE graphs. The MTIE plot shows 0.1 ms maximum error after 10k seconds and likely sub-ms error at one day.

Stable32 plots, 50k seconds of data