CDF Note #4422 (11/19/97)
Using the Cypress HOTLink Chipset to Transmit Data over Serial Links
In the Run II Muon Trigger, HOTLink transmitter/receiver chipsets are used to transfer trigger primitive data over high speed serial links. Muon trigger primitive data is formed on mezzanine cards which attach directly to the TDC cards responsible for digitizing the corresponding muon chamber or scintillator signals. This trigger data is subsequently passed back onto the TDC and sent to a transition module at the back end of the card via the J3 connector. The data transmission circuits on the transition module are designed to pass 28 bits of trigger data each 132 ns crossing interval over a single serial connection. The Cypress HOTLink transmitter chip itself is designed to transfer eight bits per clock cycle which implies that a 33ns clock cycle is required. Within each eight bit transfer a single bit is reserved for marking the position of the corresponding data within the four transfer cycles required per 132ns crossing interval.
The HOTLink transmitter chip produces a differential PECL output which can be sent over Coax, Twisted Pair, and Optical connections. All serial links originating from the TDC transition modules are routed to the muon matchbox crate which is located in the first floor counting room. The TDC cards for the central part of the muon detector are also located in the first floor counting room, and the serial connections between these cards and the muon matchbox crate have relatively short cable runs. In the case of the forward muon detector (IMU), however, the TDC cards are located in the collision hall,
and the resulting cable run for serial connections between these cards and the matchbox crate is on the order of 250 feet. We have tested two types of possible connections for use in conjunction with the HOTLink transmitter/receiver chipset. The first is an RG-58 coax connection, and the second is an optical fiber in conjunction with an AMP molded-optronic data transmitter/receiver pair (269051-1, 269052-1). This document will describe the test circuits which were built to test each of these options and the results of these tests.
As discussed previously, the HOTLink transmitter/receiver chipset is designed to transfer data over high speed serial links (fiber, coax, and twisted pair) at rates of 160 to 330 Mbits/second. The transmitter chip runs off a user supplied clock and latches eight input bits on each rising edge. Before transmission, the eight bits are converted internally into ten bits using NRZ modulation, and then multiplexed onto the differential PECL outputs. Non-Return-to-Zero is a type of data encoding which limits the DC component of the transmitted signal and ensures that multiple transitions are seen in the serial output stream during each clock cycle. The 33 ns clock period required for the Run II muon trigger is at the upper limit of the HOTLink operating range. This clock frequency corresponds to a net user data rate of 242 Mbits/s and an overall rate of 303 Mbits/s. It is important to consider the total data rate when designing output drivers and receivers.
In this application, the two important clock frequencies associated with the operation of these chips is the user supplied clock frequency (30.3 MHz) and the internal output multiplexing frequency (303 MHz). At these frequencies, the HOTLink chipset can be both sensitive to, and a source of, noise on a PCB. To ensure proper operation, this sensitivity can be reduced by providing a low impedance path to shunt transient energy to ground. This functionality is obtained by adding bypass capacitors to both the HOTLink transmitter and receiver chips. The impedance of a real capacitor is frequency dependent, and has a minimum value at a well defined frequency. Two common values of capacitors which have an impedance minima close to the above operating frequencies are as follows:
C = 56pF, fmin = 300.7 MHz
C = 5.6 nF, fmin = 30.07 MHz
The HOTLink transmitter and receiver chips each have three VCC power pins, and each should by bypassed to ground using one each of the above capacitors. These capacitors should be placed on the same side as the given power pin and as close as possible to the chip. Additionally, all VCC power pins should share the same power via. The inductive reactance of a single via provides some additional filtering of both noise generated on and off of the HOTLink chips. Similar rules apply to the ground supply pins, although via sharing is not necessary. An example layout is shown in the figure below.
The choice of RG-58 as the transmission media for our tests of a coax connection using the HOTLink chipset was based on general familiarity with the product. The chosen test configuration was a dual transformer-coupled interface. One of the primary benefits of this type of interface is that it allows for isolation of the ground planes between the PCB boards which contain the transmitter and receiver circuits. Due to a significant concern over the potential acquisition of digital noise on the detector front end, ground isolation between crates containing the transmitter and receiver circuitry is a critical requirement for any serial link which one might use in the muon trigger. In addition, transformers are a natural choice for providing the balanced-to-unbalanced conversion which allows for the transmission of the HOTLink PECL output along the coaxial cable. The inherent impedance of RG-58 is 50 Ohms and the signal at the receiver end must be terminated to this value. An additional transformer on the receiver end of the connection re-converts the incoming signal off the coaxial cable back into differential PECL with the reference voltage for this conversion being set with a simple voltage divider. The converted input is then passed into the HOTLink receiver chip which extracts eight bits of parallel output per clock cycle from the serial input stream. The circuit described above is shown in the following diagram.
Tests of the configuration described in the previous section have shown reliable performance for coaxial cable lengths up to 110 feet (33.5 meters). The test hardware was designed to simulate the configuration of the HOTLink transmitter/receiver chipset planned for the Run II muon trigger. Eight bits of random data are generated for each cycle of the 33 ns clock. On the following edge of the clock, these bits are latched
into the HOTLink transmitter chip and stored in an internal FIFO of programmable depth. The serialized data is sent out over a variable length coaxial connection and subsequently received back at the board. The depth of the internal FIFO is set so that its output can be compared with the reconstructed parallel data which emerges from the HOTLink receiver chip.
For cable lengths up to 110 feet, no errors are detected for test runs lasting several days. For slightly longer cable lengths, a small number of errors are typically seen for tests of similar duration although no loss of synchronization between transmitter and receiver is observed. Above some still to be measured cable length, synchronization between the HOTLink transmitter and receiver chips can not be maintained.
Optical fiber connections have several advantages over the coax connection described previously. First, coax cable is a conductor with finite resistance. A signal transmitted across a coax connection will suffer resistive losses and be attenuated over a relatively short distance as compared to the same signal transmitted optically across a light guide. Second, the signal wire inside a coaxial cable is not perfectly shielded so it acts like an antenna, both radiating and picking up electromagnetic noise. Finally, the design of a "signal detect" circuit (a circuit which senses the presence of a valid data stream by detecting a sufficient rate of signal transitions without loading or distortion) is fairly complicated for copper interfaces. The main disadvantage of using optical connections
is the increased cost per link due to the additional requirement of an optical driver and receiver pair.
The AMP/Lytel molded-optronic data link is one relatively inexpensive (~$120 per link) option for the design of an optical fiber connection. The transmission medium, optical fiber, is a simple dielectric light guide which is unaffected by any amount of ambient electromagnetic noise. The frequency dependent attenuation of the signal produced in the AMP/Lytel optical driver is far superior to that of the coax connection. The rating of the optical driver (269051-1) and receiver (269052-1) pair expressed as a frequency-distance product is 270 Mbit/sec x 2 km. When used to drive a 303 MHz output signal from the HOTLink transmitter chip, the maximum transmission distance is decreased to 1.75 km which is still about 20 times longer than the longest cable runs in the muon trigger. In addition, the optical receiver contains a built-in signal detect circuit which helps to simplify board design.
Although the optical receiver is more noise sensitive than the driver, noise decoupling circuits are utilized for both. The following diagrams show the bypass circuits which should be used in conjunction with the optical driver and receiver.
The low frequency polarized capacitors should be surface mount tantalum (10 uF). The high frequency capacitors should be surface mount multi-layer ceramic with dielectric material X7R, NP0, or C0G (330 pF, 1 nF). Power arriving at the optical driver and receiver is also filtered through ferrite suppression beads which exhibit greater losses at high frequencies than wound inductors. The surface mount ferrite beads chosen for this application are manufactured by Vishay/Dale (ILB-1206). A 120 Ohm resistance value for the ferrite bead is chosen to maximize attenuation at the input/output frequency of the optical driver (303 MHz). Each of these different components should be placed on the same side of the board and as close as possible to the optical driver or receiver when laying out the associated printed circuit board.
The hardware utilized to test the optical connection described in the previous section is identical to that which was described for the coax connection. The longest optical fiber tested so far has a length of 380 feet (114 m). Tests of the optical connection with this length of fiber have shown no errors in several days of running. As longer cable lengths become available, these tests will continue.
The serial links required in the present design of the Run II muon trigger fall into two separate classes. One set of links is required to bring trigger primitive signals from
IMU TDC crates which reside inside the collision hall up to the muon matchbox crate
in the first floor counting room. The length of these connections is on the order of 250 feet which means that the coax connection presented in this note would not be a suitable candidate for these links. Furthermore, there is a serious concern over any potential sources of electromagnetic noise within the collision hall, and the utilization of copper connections for these links would present problems in this regard. For these reasons, the optical connection discussed in this note is planned to be used for this set of links. The second set of links are those which bring trigger primitive signals from TDC crates in
the first floor counting room to the muon matchbox crate. Since the average length of these connections is much shorter (~50 feet), the coax connection discussed in this note would be a suitable candidate. However, there are several reasons for considering using optical connections for these links as well. One argument is that one would like to have uniformity within the entire muon trigger. If both coax and optical connections are used in the muon trigger, multiple types of transition boards will be required. The extra cost
of optical links is somewhat offset by the need for only a single transition module design. In addition, the fact that the muon matchbox crate will receive on the order of two hundred serial connections from different sources implies that noise radiated and picked up between different links could be a potential problem. The use of optical connections for all of the links within the muon trigger would provide some additional insurance against this type of problem. Due to these factors, the present baseline design for the muon trigger calls for optical connections in all serial links within the system.