GPIB to USB controller/adapter

[free_electron]

Possibly.
GPIB is very fast synchronous bit parallel, byte serial bus with handshaking and multimaster capability. HP actually had computers where the harddisks , floppy or tape drives were connected on the HPIB bus. The drive unit had its own controller. See it as an early kind of SCSI.

Those controllers used DMA techniques to tranfer large blocks of data very efficiently.

The reason GPIB controllers cost money is due to licencing.
HP's licencing model has always been
-devices do not need to pay licences. Meaning, everyone is free to make a piece of testequipment , or other equipment, that can connect to a GPIB link as a device.
-controllers need to pay a licence. So any computer or instrumnet with those capabilities (instruments that can drive a plotter for example) need to cough up some dough to HP.

That is the reason those boards are so expensive.

Making the device usage free guaranteed a very quick uptake by other manufacturers so they would be compatible. Since you have multiple devices on a bus the cost of the host is shared amongst devices.

National sells their TNT4882 asic in two flavors : one that is host capable , one that is not. They fry an internal fuse. The standard chip is like 4$, the host capable one requires signing of paperwork and costs 54$...

Traditional chips have now all gone the way of the wind. TMS9914, upd7210 , MC68488, hef4788, intel 8291/8292/8293 are all gone.

The intel solution was actually a really nice one. 8293 was their equivalent for the 75160/75161 transceivers. The 8291 was the base GPIB logic. Add on the 8792 to become a master. So only the 8792 requires licencing. It was actually a mask programmed 8048 processor.

National still sells the NAT7210 and NAT9914 as drop in replacements for the NEC and TI , but i think the master capabilites are disabled in those.

Agilent is facing problems in their instruments as well as the original devices are gone. They went so far as to implment the logic either in the ASIC, and for older machines they had a custom chip made by TI called the ML9914. This is again a 9914 that has no master capability.

[mrflibble]

So in summary the  SN75160B + SN75161B are the only "proper" ones left that are actually available and affordable? As for ethernet, I'd say 10 MBit is perfectly fine for just controlling and doing regular readouts. If you need lots of binary transfers then 100 MBit should do the trick in most cases. Gbit is nice but would seem bit overkill to me, especially if you have a nice long GPIB cable dangling on the other end. Parallel port transfers with dma (ECP and all that) pooped out round 35 MByte/s, and that was with a short cable. Don't see GPIB with it's open collector and long cables and multiple devices attached going to reach stellar speeds.

Anyways, that's from a homebrew perspective. 100 mbit is easy enough, gbit not so much. If you are going to professionally manufacture those might as well go for gbit to get that extra bit of transfer speed + lower latency.

[free_electron]

SN75160B + SN75161B are only the bus drivers. they contain no intelligence.

With all those chips i listed ( apart from the intel ones ) you needed the SN75160B + SN75161B as well.

those bus drivers are special. they can drive easily 50 to 100 mA per pin.
GPIB is NOT open collector !
the TALKER is alsways driving the bus hard from a totem-pole output. If a device is on the bus but in off-line state ( not listening , meaning not beeing addressed) then
it is in open-collector mode with pullups off. So it forms no load on the bus. This is why a device that is powered off has no impact on the bus. you can leave it connected . it does not disrupt bus activity

A connected device in listening mode (being addressed ) turns on its pull up resistors.

so essentially the pull up resistors on the network are always where they need to be : at the device that is the current active listener. This guarantees good signal integrity without having to handle the load of 10 pullup resisotr sin parallel if 10 devices are connected.

GPIB is a very cleverly constructed bus.

[MarkL]

The Ethernet side only supports 10 Mb ethernet, but that's already faster than the GPIB side so it's not really an issue. 

you need to read up on your standards ...

1)  10 megaBIT ethernet has an effectiveusable rate ( if on a dedicated link ) of about 9megabit ( there is overhead bytes in mac and IP address. The usable payload is smaller)

2) GPIB in its base 488 standard is a 10 megaBYTE transfer rate and can go 25 Megabyte/S in 488.2 (Turbo mode). you need at least a 100base-T link to keep up with a GPIB bus going flat out.

As per IEEE 488.1, interlocked GPIB transfer rates top out at about 1Mbyte/s.  NI introduced non-interlocked transfers (aka HS488) in IEEE 488.1-2003, which has a maximum rate of about 8Mbytes/s and depends on the total length of the bus.  I've never heard of "Turbo" mode (is that some proprietary implementation?).  488.2 specifies higher layer command structures and doesn't deal with physical layer transmission rates and timing.

GPIB is NOT open collector !
the TALKER is alsways driving the bus hard from a totem-pole output.

Actually, 488.1-2003 (sec 5.3) permits the use of open collector drivers and requires it on certain signal lines.  It also requires it on the data lines if parallel poll operations are to be performed.  However, using open collector drivers during regular data transfers have a detrimental impact on the bus speed due to the slower rise time, and three-state drivers are therefore recommended.

To the original question on interfacing GPIB to Ethernet, most pieces of equipment have slug-like GPIB implementations and wouldn't get close to saturating a 10Mb Ethernet.  You will find HS488 on NI's controllers (no surprise there) and some Tektronix gear, but I don't think it was widely implemented by most equipment manufacturers.  I wouldn't feel that 100Mb Ethernet had to be in the mix for a workable solution, let alone 1G.

[free_electron]

Oops, now i got my revisions confused. 488.1. Is hardware, 488.3 is software.
Turbo mode is a mode implemented by the tnt4882 asic . This can go. Uch faster than standard gpib ,but only for data transfer. Timing of the other signals is the same. It uses a 48MHz master clock and data is pumped at half that speed.
This mode only works between the tnt4882 chips. It is national instruments proprietary.
I had their development kit at one time. We built a machine with that chip in and could use that mode to stream samples very fast.

As for the open colllector/ totempole discussion: of course the drivers switch to open collector during a parallel poll as you do not know how many responders will drive.
The point is you need a chip that can switch between totempole and open collector. Sinply i plementi g a dumb open collector only with pullups is going to give trouble, especially on long lines and with 4 or more instruments attached. I have setups with 20 or more. Machines where i need bus repeaters to improve drive strength. The standard 75160 can handle 15 devices without problems.

[macboy]

SN75160B + SN75161B are only the bus drivers. they contain no intelligence.

With all those chips i listed ( apart from the intel ones ) you needed the SN75160B + SN75161B as well.

those bus drivers are special. they can drive easily 50 to 100 mA per pin.
GPIB is NOT open collector !
the TALKER is alsways driving the bus hard from a totem-pole output. If a device is on the bus but in off-line state ( not listening , meaning not beeing addressed) then
it is in open-collector mode with pullups off. So it forms no load on the bus. This is why a device that is powered off has no impact on the bus. you can leave it connected . it does not disrupt bus activity

A connected device in listening mode (being addressed ) turns on its pull up resistors.

so essentially the pull up resistors on the network are always where they need to be : at the device that is the current active listener. This guarantees good signal integrity without having to handle the load of 10 pullup resisotr sin parallel if 10 devices are connected.

GPIB is a very cleverly constructed bus.

This is not correct, GPIB is definitely driven open-collector. Only with "newer" ( ) high-speed devices is the bus actively driven High. Absolutely all devices must conform to the open-collector mode of operation and optionally can use what HP calls tri-state, where it can drive high or low or go high-impedance for input. The SRQ, NDAV, and NRFD are all open-collector only.  In either case, the receivers need to conform to (inverted) TTL levels: 2.0 V or higher is a 0 and 0.8 V or lower is a 1. This is where a lot of homebrew implementations fail; people use CMOS microcontrollers that need 0.8*Vcc (4+ V) to register a reliable High level (and which at 5 V Vcc will usually register < 2.5 V as low, contrary to the spec).

Every low-level detail of the GPIB can be found in HP's "Tutorial Description of the Hewlett-Packard Interface Bus".

[MarkL]

This is not correct, GPIB is definitely driven open-collector. Only with "newer" ( ) high-speed devices is the bus actively driven High. Absolutely all devices must conform to the open-collector mode of operation and optionally can use what HP calls tri-state, where it can drive high or low or go high-impedance for input. The SRQ, NDAV, and NRFD are all open-collector only.

Well, that's not quite right either.  Here, this is what 488.1-1987 has to say about it (and it's identical in 488.1-2003):

3.3.1 Driver Types

Open collector drivers shall be used to drive the SRQ, NRFD, and NDAC signal lines.

Open collector drivers or three-state drivers may be used to drive DIO 1-8, DAV, IFC, ATN, REN, and EOI signal lines with this exception: DIO1-8 shall use open collector drivers for parallel polling applications (see 2.9.3.3 ).

NOTE — Three-state drivers are useful for systems where higher speed operation is required.

It is recommended that a three-state driver be used within a controller to drive the ATN signal line if the controller is intended to be used in a system in which other devices are implemented with three-state drivers on the DIO, DAV, and EOI signal lines.

Every low-level detail of the GPIB can be found in HP's "Tutorial Description of the Hewlett-Packard Interface Bus".

The actual spec works pretty well too.

[free_electron]

This is not correct, GPIB is definitely driven open-collector. Only with "newer" ( ) high-speed devices is the bus actively driven High. Absolutely all devices must conform to the open-collector mode of operation and optionally can use what HP calls tri-state
[/quote]
All the boards i know of (and there are a lot) use totem pole drivers ( what is called three-state ). When a device is talking the drivers are switched in push-pull and the listener has its pull ups enabled. When a device goes off-bus it disables its pull ups.
The 75160 and 161 has a special io structure to make sure that there is no load when the chip is without power. you can turn off a device ( cut power) while leaving it attached to the bus. This has no impact.

I've done a number of custom instruments that had a GPIb interface using both the uPD7210 and the TNT4882 chip.
I have seen implementations 'el-cheapo' where they hard strap the control pin on the 75160 so it becomes an open collector device but that is NOT the recommended way.

[xzswq21]

I have an old HP spectrum analyzer (HP8560), so I need a GPIB to USB adapter to capture some pictures from the Screen, there are many GPIB to USB adapters in eBay and I'm confused to choose one!

Prologix:
https://www.ebay.com/itm/Prologix-GPIB-USB-Controller-/265103834697?_trksid=p2349624.m46890.l49286

Could you please help me to find a proper Adapter in eBay? Not expensive. I don't have any problem with the used original adapters. Thanks