I recently broke down and purchased the UniSource FC-8300: a professional, 3.7GHz, 9-digit frequency counter, which I obtained for $312 from TestEquipmentDepot (
http://www.testequipmentdepot.com/unisource/frequencycounter/fc8300.htm). I just thought I'd share my thoughts on this unit so far.
This device is actually from a reference design which is also sold by several other companies, including B&K Precision (model: 1856D), and EZ-Digital (model: FC-3000). These are the exact same products right down to the User Manual, and as such the software (for remote access) is cross-compatible between vendors. Connectivity is achieved via RS-232, and example code is supplied along with a demo application which allows the user to set the operating mode, state, gate time, and other features remotely, as well as display the value that is being read on the screen. The UniSource software graphs out the frequency values for you in real-time, whereas the B&K Precision software allows this data to be stored to the hard drive in the form of an Excel spreadsheet.
According to the datasheet, the default temperature compensated oscillator (TCO) has 1ppm accuracy (not including aging) and this is fairly accurate as measured using my FE-5680A Rubidium Frequency Std. The one gripe that I have with this unit is that while it does support an external 10MHz input (as any good counter should) via a switched BNC connector on the back, it only accepts inputs that are 1.5 - 5 volts RMS. Without external circuitry, my rubidium oscillator only puts out 0.7 volts RMS. There is also apparently other options available for the internal timebase (such as ovenized crystal); however, I could not find any information about ordering these options.
This counter features two inputs: a DC-coupled BNC (Input-A) which goes up to 100MHz, and an AC-coupled (Input-C) N-female which goes from 80MHz to 3.7GHz. Input-A has a 1Mohm input impedance and can safely accept input voltages of up to 200 volts peak-to-peak. Input-C (I have no idea why it's not called Input-B) is terminated with a 50-ohm impedance and can only accept 3 volts on the input (about 20 dbm). I have tested Input-A to its full bandwidth and Input-C up to 1.5GHz and it is accurate throughout these ranges.
As far as features are concerned, this unit has three modes of operation: frequency (uses reciprocal method (auto) for low frequencies), period, RPM, and Total. Note that only the frequency mode is available for Input-C. The gate time can be toggled between 10ms, 100ms, 1s, and 10s (an extra digit of resolution is made available to you with each increase in gate time). There is a Hold button to freeze the display, a button for manually enabling AC coupling on Input-A, a button to enable a 100kHz low-pass filter to reduce noise when measuring low frequency signals, and the input attenuation can be selected to 1X or 10X.
I did open the case of this counter to view the inner construction quality (you all know Dave's motto) but unfortunately I didn't have my camera with me at the time. What I can say is that it looks very nicely constructed with no bodge wires or cold solder joints, and the entire PCB is enclosed in a metal shield to prevent external EMF from affecting high frequency accuracy. The main microcontroller is an Atmel 40-pin IC in a DIP package, and there are plenty of other dedicated ICs for handling the work. I was afraid for a moment when I realized that one of the wires leading up to the Ext. reference was disconnected (I thought it broke off), but it is actually not connected by design (I tested the ext reference and it does work properly). The fuse can be accessed via the rear panel, as can the mains voltage switch (120/240). There is also a banana jack receptical for Earth GND (in case anybody needs that), and the RS-232 port as mentioned above.
Here is a photo of the front of the device for your enjoyment.