Hattix.co.uk PC Hardware Collection
     
  NICs, Modems, Networking
US Robotics Sportster 33.6 #840 modem (V.34) - Late 1996

You now have the ITU dialup handshake tones (or "that abominable screech") in your head. The initial buzzing noise was frequency shift keying (FSK) data where the modems started out at V.8bis, upgraded to full V.8, disabled the echo suppressors on PSTN (or tried to), then told each other what they could do. Just after that, and before data began, a high pitched series of tones measured the telephone line and then agreed on a working symbol rate. They then trained against each other and began proper user-data transmission. The speaker is muted at that point.

Back in the day, there was a big thing about "real hardware modems" against "Winmodems". Winmodems implemented the DSP in software, and the modem itself was just a digital to analog converter and a telephone line interface. Hardware modems did not, they ran off a serial interface (or a fake one over a PCI or ISA slot) and handled all the data compression and encoding themselves. In the late 486 and early Pentium days, a Winmodem could substantially hit the CPU. By the time CPUs reached 200 MHz and beyond, it was a non-issue.

This one, like quite a bit more around here, came via Bryan McIntosh:

It's a US Robotics Sportster 33.6K modem, which I received as a birthday present from my parents in January of 1997. It was supposed to be my gateway to the Internet even though I was using a crusty 386 DX-25 at the time with Windows 3.11. We never ended up actually connecting the modem to anything since my parents (wisely) insisted that we get a second phone line before I started inevitably tying it up while surfing the mid-90s web; we got the second phone line in the Fall of 1998 around the same time that I got a shiny new PII-300, my first modern computer (more hardware from that box to come when I get around to taking pics), and this Sportster just sat inside the 386's case waiting to be used.

Although it's a V.34 modem supporting a maximum of 33.6 kbps, this modem had another odd party trick up its sleeve; USR had a technology that they called "x2" where for a nominal fee, they would upgrade your modem [Wayback Machine link] to a 56K modem! This is likely what the socketed chip is for; they'd likely pop that out and replace it with the controller chip for a 56K x2 modem.

Of course, the x2 standard fought against the K56Flex standard, and you had to make sure that your ISP supported your modem's 56K technology, otherwise you were stuck at V.34 speeds. Eventually most ISPs that offered 56K service switched to the V.90 standard, and I imagine that modems like this one were left with a useless 56K technology. So, this card has been installed, but has never actually been connected to a server anywhere in over 22 years!
A look around USR's legacy site finds how we can work out what model we have: It's the SKU 000840-03, identifying it as a Sportster 33.6 #840. Of course this doesn't much help, as there were eight versions of the 840, some compatible, some not.

Had we blown the $60 for the X2 upgrade in 1997 or so, USR would have sent out a PLCC chip removal tool and a new PLCC chip to fit on the modem. This would have got a maximum sync rate of 53,333 bps (in real life, this would have been between 40K and 47K). Rockwell sponsored the K56Flex standard, while USR developed the X2 standard, neither of which were open standards, and neither compatible with the other. In general, K56Flex saw more support due to ISP equipment more easily being upgradable to the K56Flex standard: Rockwell made nearly 80% of all ISP modem equipment.

US Robotics was a well regarded vendor, and this is a well implemented device. We'll have a vaguely clockwise tour of the PCB, starting at the upper left.

That big round thing with the holes in it is a speaker! It's responsible for the modem handshake and training sound. Moving on is a little 28 pin PLCC package, labelled TLC 320AC01C-FN, an analog interface. Today we'd call it a codec, as it is integrated full-duplex ADC and DAC. It has an internal low-pass filter which cuts at 3.6 kHz. It feeds the speaker, via either the op-amp just above it or the one just below the speaker, but is primarily responsible for interfacing with the telephone line.

The signal going into the analog interface goes via the big yellow thing, a 1:1 isolation transformer. The beige/brown/orange thing is a metal-oxide varistor, which protects against line surges. It lacks a ring capacitor, which regenerates the ring signal on a phone line chain, so the device along in the chain, if an old analogue phone, would not ring (these were extremely rare by the time of modems). The blue thing beneath it is a relay. This "picks up" the line.

Being 8-bit ISA, this guy needs jumper configuration, and the jumper block allows us to configure which COM port and which IRQ the device will be configured to use. It can use IRQ 2, 3, 4, 5 or 7.

Just to the left of this is the TL16CFM504APJM controller, which is a member of the TL16C ISA controller family. They are dual UARTs with 16-byte FIFO buffers. This handles the ISA bus and acts as the first stage of an ISA to RS-232 converter.

Just next to it is a National Semi 74VHCU04 hex inverter (six NOT gates) just above a 4.281 MHz resonator (an oscillator has four legs, a resonator has two) which isn't a standard or common frequency. Moving along, the 80 pin PLCC is a TI PD17804PJ CDSP (custom DSP) which is linked to the two chips below it (they're in parallel), which come later. The 80 pin package is the main modem DSP. The two identical chips below it are SRAMs manufactured by ISSI: The IS61C256AH is a 32,768 word, 8 bit CMOS static-RAM, so 256 kbit or 32 kByte. These have date code week 37, 1996 and, together, they give a 16 bit 64 kB SRAM capacity, used by the DSP for workspace ("scratch"), buffering, etc.

Finally, the Atmel AT27C2048 (date code week 41 1996) is a 128Kx16 (2,097,152 bits) OTP EPROM, programmable at 12.7V but only one-time programmable. Faint markings of perhaps a sticker can be seen on the EPROM, which appears to be a copyright notice and a date. The replacement, for X2 operation, had a fancy X2 logo on it, so we know this one wasn't upgraded.

Realtek RTL8029AS
ISA ethernet network adapters were usually BNC and later RJ45 for 10BaseT, which rapidly emerged as the dominant medium. It was, however, quite rare to see 10BaseT in both BNC and RJ45 forms on a PCI card, this arrangement normally reserved for their elderly ISA brethren.

The RTL8029 was one of the few 10 megabit adapters with PCI compatibility, and also supported 10Base2, 10Base5, however neither could handle hot-plugging the network cable.

As this was at the end of 10BaseT's life, there were many features on-die. It could program a 9346 EEPROM without any assistance to use as boot-ROM. Like most of the day, it was NE2000 compatible, like this really meant anything, it could work in a Novell Netware environment.

The Novell NE2000 was Novell's way of selling more Netware licences. It was a crude National Semiconductor prototype (8390), but Novell could not sell much Netware since IBM and 3Com charged an absolute fortune for networking hardware. Novell decided to produce a reference design around National's 8390, the NE1000, for $495 USD in 1987. This undercut 3Com by about 50%. There was no royalty needed for compatible designs, and Novell allowed its distributers to go direct to the manufacturer. AMD's (buggier than an ant farm) LANCE-7990 was a significant competitor.

This RTL8029AS was originally intended for Cat-3 cable, as can be seen in that only two pairs are used, it wasn't until Cat-5 that four pairs of conductors were used. UTP cable allowed full duplex operation, and although the RTL8029 wasn't great at it, at least traffic in one direction did not cripple traffic in the other.

The 8029AS was basically an 8019AS with a PCI interface.
Intel 536EP 56k PCI Modem
Intel had long believed in using the power of their CPUs wherever possible and their modems were no exception. The 536EP was a development of Ambient's PCI softmodems, after Ambient had been bought by Intel. The main DSP was implemented in hardware, but the error correction and compression was done in software, making it a hybrid of hardware and software, not a true hardware modem and not a true software modem.

As we can see, this uses an isolating transformer between the digital and analog sides of the modem, it was before silicon DAA isolators were common, though it does have a pair of MOVs (the gold coloured components near the line sockets) to protect against transients.
Holy Crap, It's an AMR card!
AMR slots were the Intel version of the more open CNR riser and were exactly the same thing. They appeared on quite a few motherboards but it was exceptionally rare to find them in use and the cards themselves could be bought from virtually nowhere.
They are simply the analog side of a dialup modem, the digital side done in the MC'97 southbridge and on the host CPU. In this one, like all dialup modems, the blue components you see are metal oxide varistors (MOVs) which are commonly used for surge protection, the big brown thing is a non-polar high voltage capacitor (250V rating) used to regenerate the ring line on a UK (BT) line, where the ringer can be regenerated - this modem had a line pass-through. As the ring line is a third line and RJ11 carries only two conductors, it is regenerated on each device.
The upper IC is a (PCTel) PCT303W DAA (data access arrangement) phone-line interface providing isolation and the lower is a Silicon Laboratories Si3024-XS8 DAA which interfaces with the MC'97 southbridge and converts the MC'97 serial audio data back into analog audio for either voice or data use.
A similar pair of DAA chips in 16 pin SOIC packages can be seen on most modems, common pairs are the PCT303W/PCT303D and Si3012/Si3024 (collectively known as the Si3036 - datasheet). Modems without DAA ICs have to use isolating transformers and seldom come with MOVs.
Realtek RTL8139A
A cheap, no-nonsense NIC bought in late 1999 and still in daily use in 2007. This particular specimen cannot run reliably above a 35MHz PCI bus, so facilitated zero FSB overclocking. For that reason, it was demoted to the dual system which can keep its PCI bus in spec at its normal working frequency. The activity LED was once cut out and the wire soldered on to work a front panel LED. Sadly, the network it was on at the time was hub based and not switch based, so the LED was always flashing.

The RTL8139 itself was never a great performer, being about as basic as a NIC could get in terms of function, so needing the host CPU to handle pretty much everything. However, the RTL8139 was cheap and very reliable so it made design wins across all the top-tier OEMs; An 8139 just works without needing external drivers on everything since Windows2000 and up. Many techs carried 8139s in their toolkit, such is the importance of network connectivity and if the card got broken it was cheap and easy to replace it.

The "A" variant denoted it had Wake-on-LAN functionality via "magic packet" or "wakeup frame", but it needed an extra cable to the motherboard to enable this. It needed split signal for the clock, 25 MHz and 50 MHz. This was handled by the 50 MHz oscillator (position X1) and the small 8-pin package beneath it.
Realtek RTL8139C
This one is similar to the above, but on a larger PCB and using a later revision of the 8139 controller. It can handle an overclocked PCI bus up to 44MHz. Neither of the two Realtek NICs are great performers but they can push 80Mbps when asked to.

The "C" version is made for 3.3V operation and has ACPI compliance to PCI 2.2 specification, so does not need a WOL cable for Wake-On-LAN. It includes the "B" variant's ability to work from a small 25 MHz crystal oscillator (position Y1). The "D" revision (not terribly common) re-adds 5V compatibility.
Realtek RTL8139C
Genius are known for end-consumer marketed upgrades (usually keyboards, webcams and mice), which means their quality somewhat sucks and their price tends to be high; but the box and packaging tends to be flashy. This is about as cheap and trashy as an 8139C could ever be. This card cost twice as much as the one above, even though it's obviously cheaper to manufacture and was three years later. It never performed reliably (which was incredible for an 8139) and eventually was retired from service to be replaced by a Netgear.
ZyXEL Prestige P660R-61C

Well built to good quality standards, ZyXEL make good workhorse SOHO network products. It is based around the Texas Instruments AR7 single chip DSL modem/router using the 7300A chip. The firmware is stored on Intel flash memory (2MB capacity) and no custom firmwares are available due to ZyXEL's habit of using specially modified bootloaders and a custom OS, ZynOS. The 2MB capacity somewhat puts a damper on things too.

The router does not have the ability to set target SNR and, indeed, most commands on the CLI are locked out and do nothing. DMT5.22 will work with this router, but little else will.

UPnP is supported but with UPnP enabled most games will suffer great lag until the router is rebooted. Running BitTorrent clients seems to cause this and, regardless of whether the client is still running or not, only rebooting the router will solve anything.

It's cheap, it's well built, it's easy to configure and use, it usually just works without issue, but it's not for the advanced user or anyone who wants to be hacking firmwares.
Realtek RTL8169S

I couldn't get the heatsink off without potentially damaging the card, so it remains unidentified. There's no UL or FCC number on it and the only identifying mark on the back is a barcode sticker with "NC-1000TX-R06030973" printed on. My best guess from the size of the IC and layout of the card is some variety of Realtek, possibly an RTL8169 or RTL8110. As it has a heatsink, it's pretty early, so I'll tip it to be an RTL8110.

A single port PCI NIC with a heatsink, whatever next?

Update - It's been put into use and identifies with a vendor ID of 10EC, which is Realtek, and a product ID of 8169, which likely puts it in the 8169/8110 family. However we need to be cautious about a firm identification. While they're largely software compatible they do differ in their product IDs. Realtek have long used sensible numbers in the PCI ID codes, but a RTL8110SC I have here identifies as 8167. They use an identical driver, right down to the same files, and appear identically in Task Manager. This card doesn't have J1 (WOL connector) fitted and it knows it; Properties page in Task Manager omits the Power Management tab.

Update: With the heatsink finally off, the chip is clearly seen to be an RTL8169S, one of Realtek's first ventures into gigabit ethernet. Due to the complex DSP required to operate on upwards of 100MB/s (not that PCI can sustain that), these first generation models all required small passive heatsinks. Judging by its peers, this heatsink is smaller than the others, so may well be the cause of its slight instability.
Anatel RT3090 PCIe WiFi Card

This little guy lived in an extremely unlovely Packard Bell imedia S1800, manufactured March 2011. The date of manufacture on this card is Feb 14th 2011, the HDD (A 7,200 RPM 1 TB Hitachi, probably the best component in there) was October 2010. A Valentine's Day WiFi card, is there anything to love about it?

No. No there is not.

It uses the entirely basic Ralink RT3090 chipset, supports the usual 802.11b/g/n (initial certification was draft-N), doesn't support the 5.2 GHz band, doesn't support MIMO and so can peak at just one spatial stream of 40 MHz, so a peak symbol rate of 150 Mbps. This would translate to a data rate of about 80 Mbps. It wasn't even any good for the time. It did support antenna diversity (two antennae) but only in so much as switching between them for the best signal, it wasn't in any way able to do MIMO. Either way, the second antenna feed is not fitted at all on this card. It isn't even an unfitted option on the PCB!

The design is our next clue. This cut every corner possible, then more. Usually, a desktop WiFi adapter would be a mini-PCIe thing intended for laptops, placed on a PCIe x1 slot adapter. This, if you're wanting decent WiFi, is the most cost effective, as mini-PCIe cards are made in immense bulk and the adapter is no more than a mounting and the fitting for the port. Dell does this across its entire range.

However, if you're a decently sized OEM, like Packard Bell, you can specify whatever you want to keep costs down, if you're ordering enough of them. If you're cutting costs everywhere else (the sole video output on this system was VGA, when VGA was obsolete already!), half a buck on the WiFi card soon adds up.

Crummy card from a crummy PC, but installs out of the box on Windows 7 and up.

     
Script by dutches   © 2002-2018 Hattix.co.uk. All rocks deserved. Creative Commons LicenseHattix hardware images are licensed under a Creative Commons Attribution 2.0 UK: England & Wales License except where otherwise stated.