Adapter module, RJ45 to 8 SMA Female

APPLICATIONS

• Transmitting multiple LVDS, RS422, NECL or other differential signals over long runs over inexpensive cabling
• Reducing cable count
• 1000BaseT and 10GBaseT Break-Out/Break-In¹
• Ethernet signal probing and characterization
• Signal Injection for Test

FEATURES

• Shielded RJ45 jack
• 8 SMA jacks
• TIA/EIA-568B or TIA/EIA-568A pin mapping
• Matched 50 Ω impedance traces, suitable for adapting 100 Ω differential pairs

The PRL-RJ45-SMA adapter connects 8 SMA circuits to standard Cat 5e (or better) cabling.

The center pin of each SMA is connected to a signal pin on the shielded RJ45 connector, and the SMA shields and the RJ45 shield are all tied to chassis ground. Each top/bottom pair of SMAs is connected to a twisted pair in the Cat 5 cabling, based on the common TIA/EIA-568B or TIA/EIA-568A wiring, e.g:

The PRL-RJ45-SMA is designed primarily for converting differential signaling between SMAs and shielded twisted pair, but it can also be used for single-ended signaling by tying one of the SMAs in a pair to GND via an SMA termination.

If the drivers and receivers are DC-coupled circuits, shielded Cat 5e or better cabling is recommended whenever the driver and receiver are separated by long distances and/or when they are supplied by different power circuits (e.g. any time the grounds on opposite ends of the Cat 5e cable are subject to ground potential differences).

A discussion of impedance matching through this adapter (and other, similar adapters) is presented here.

The PRL-RJ45-SMA can be used with a variety of applications, such as:

• Carrying generic differential signals across inexpensive network cabling. The 100 Ohm nominal differential impedance of UTP/STP is a good match for pairs of 50 Ohm coax.
• Carrying single-ended signals across inexpensive network cabling. While not formally specified as such, the single-ended impedance of the STP wiring in Cat6a/Cat7 is acceptably close to 50 Ohms for many applications.
• Adapting true Ethernet signals from SMA connectors (such as from an FPGA evaluation board) to Cat5e/Cat6a/Cat7/Cat8 cabling, at speeds up to 10 Gbits/sec.
• Measuring, monitoring, and debugging Ethernet applications.

Generic Differential Signaling

A discussion of impedance matching through this adapter (and other, similar adapters) is presented here:

Carrying Single-ended Signaling

The PRL-RJ45-SMA is a surprisingly good match for single-ended, 50 Ohm signaling, as shown here:

TDR→Coax→PRL→RJ45→SMA→Cat7 cable, 10’

Adapting Coaxial Ethernet

Several customers have reported good success using the PRL-RJ45-SMA adapters to adapt non-RJ45 Ethernet interfaces, such as SGMII or RGMII from an FPGA evaluation board, to RJ45 for connection to standard networks.

Although PRL has not done any detailed characterization of signaling with respect to Ethernet standards, we have performed end-to-end testing that demonstrates successful carriage of 1000BaseT signaling:

First, we connected a pair of PRL-RJ45-SMA modules with a set of eight 50 Ohm coax cables, and the connected the RJ45 sides to a PC's Gigabit network interface card and to an off-the-shelf, unmanaged Gigabit switch via Cat 5e cables:

Cat5e→PRL-RJ45-SMA→8x0.5mCoax→PRL-RJ45-SMA→Cat5e

The link negotiated 1000BaseT every time, and iperf reported typical throughput of 940 Mbps, which is the same that it reported for a link via a single Cat5e cable:

Measuring, Monitoring, and Debugging Ethernet

Next, we inserted a pair of PRL-860Q-SMA differential pickoff tees into the middle of the coax section:

Cat5e→PRL-RJ45-SMA→8x0.5mCoax→2xPRL-860Q→8x0.5mCoax→PRL-RJ45-SMA→Cat5e

with the Pickoff ports connected to a Tektronix CSA8200 Sampling scope:

The extraction of 5% of the signal power did not affect the link at all, and iperf continued to report throughput of 940 Mbps:

Now we can capture the Ethernet waveforms on the oscilloscope. First, we swapped out the Gigabit switch for an eval board for the Microchip KSZ9031RNX transceiver chip, and this board has a test point where I can extract the 125 MHz TX Clock signal. I connected the Ethernet jack of the Microchip board to a PC, via my cabling-and-adapter setup. Then I used the PC to set the speed and duplex, and brought up a link.

We needed a clean copy of the 125 MHz Tx Clock signal for the scope trigger, so we tapped into the Tx Clock pin on the Microchip eval board and buffered it with a PRL-350TTL comparator module.

We connected a Tx signal and an Rx signal to the oscilloscope and captured some traces. 100BaseT is easier to handle than 1000BaseT because:

1. 100Base-TX is transmit or receive per pair, and uses 2 pairs to achieve full duplex.
2. Therefore, if we trigger the scope on the TX Clk of the Microchip board, we should be able to see clearly the TX signal from the Microchip board.
3. But we would _not_ be able to see clearly the RX pair, because that's clocked by the TX clock of the PHY on the other end of the 100Base-TX link, and that 125 MHz is asynchronous with this 125 MHz.
4. Which pair becomes Rx or Tx depends on the Auto MDI-X negotiation, so we have a coin flip's chance of getting the TX pair on Scope Ch. 1 (green) every time the link negotiates.
5. The signal has 3 discrete levels because 100BaseT uses 4B/5B and MLT-3 encoding to transmit 4 bits every 5 symbols

While we would like to show PAM-5 eye diagrams from 1000BaseT, this is significantly more difficult, because 1000BaseT is full-duplex per pair, relying on echo cancellation in the PHY to separate the Rx signal from the Tx signal, and the post-processed analog signal is not physically accessible outside the PHY in most implementations. Therefore triggering the scope and capturing a representative waveform of Rx or Tx is challenging.

The MLT-3 waveforms from 100BaseT are still a good proxy for signal fidelity measurements, because 1000BaseT uses the same 125 MBaud signaling rate at 100BaseT, relying on twice the number of pairs, 2.5x the number of bits encoded per symbol, and full-duplex per pair to increase the throughput to a Gigabit per second.

10GBaseT

We recently tested a pair of PRL-RJ45-SMA adapters with Marvell-based 10GBaseT Network Interface Cards, and they successfully negotiated a 10Gbit/sec link and carried iperf traffic at near the theoretical maximum 9.5 Gbit/sec:

The command on the server was "iperf3 -s -B 192.168.50.2", where the -B option is used to bind the command to a particular Ethernet address, since both cards are installed in the same machine.

The command on the client was "iperf3 -w 4M -c 192.168.50.2 -B 192.168.50.3", where the "-w 4M" option sets a larger TCP window, as the default 256 kB will not saturate a 10GBaseT link with a single thread. 

SPECIFICATIONS (0° C ≤ T_A ≤ 35° C)*

Notes:

1. 1. Although empirical testing has been successful, PRL can not guarantee 1000BaseT or 10GBaseT performance, as these standards do not specify performance through coaxial cabling.
2. 2. T_PROP and T_SKEW measured by reflectometry of TDR pulse.
3. 3. Measured by driving differential TDR pulses through two PRL-RJ45-SMA units and a 3’ Cat7 cable, then calculating the contributed t_r/t_f of one unit. Includes unknown t_r/t_f contribution of Cat7 cable.

While we believe these models to be accurate, no representations are made as to accuracy or suitability for any application: